Part 2: th-cam.com/video/VSm_7q3Ol04/w-d-xo.html Watch Monty's full video here: th-cam.com/video/UqiBJbREUgU/w-d-xo.html Thanks to Monty Montgomery and xiph.org for making this information available with a Creative Commons License!
Some interesting insights into DAC reconstruction filters can be seen in the two application notes by Analogue Devices AN-823 and AN-837 for Direct digital synthesis applications but theory is similar. A video describes the same details th-cam.com/video/dD9HC1GThZY/w-d-xo.html The original video featured here left out the reason that the stair step waveform is not visible on the output. It is because of the reconstruction/output filter after the DAC. Denying the existence of the stair step is a little disingenuous though.
AudioUniversity, huh?? I’ve lost count at this point of the misinfo provided as opinion, and downright incorrect info offered as education. Too bad TH-cam just doesn’t care. But I do. I now know this channel to be a safe space for cr*p, home to clickbait nonsense.
@nicksterj Were you directing the question at me? The glossing over of the fact that the output from the DAC is a stair-step before the reconstruction/output filter which is sometimes digital but traditionally analogue. The lolli-pops only exist on paper, in the digital realm it is just values. Once it is electrical it is a continuous curve (voltage or current) that will be slew rate limited but it will not be discontinuous. The stair-step is the electrical output you expect from an unfiltered DAC, trying to promote that it does not exist is not correct.
Something not mentioned is that the raw DAC is followed by a low-pass filter at the Nyquist frequency. So, even if the stairsteps are there, it's at the sampling frequency and would get filtered out.
And you filtered out the high frequency sound, too. But anyway with powerful headphone all days long can't hear them anymore. a 44kh give you the capacity to produce 22kh square wave sound. To produce decent 22kh sine wave, you need a sample a least 8 time higher. You can use Bezier algo to reshape the sound wage, but it not sound an accurate sound reproduction anymore.
@@omegaman7377 you have that backwards. With sine waves as your basis function it's impossible to perfectly model a square wave without infinite sine waves. The Nyquist frequency is plenty good for sine waves because it IS the basis function of the model. You can apply more advanced techniques to reconstruct beyond Nyquist, but not in a general way, so it's not useful for audio applications. We can do this in image processing (though with some additional information which is kinda cheating)
@@omegaman7377 Square waves have higher order harmonics, technically infinitely high, compared with their fundamental frequency. A pure sine wave is ONLY the fundamental, so you have your sampling idea completely backwards. Sine waves require lower frequency sampling to recreate without any assumption about the input signal, because you only have to reproduce the fundamental.
There's a reason to use 48khz in movies/videos which has to do with the fact that it's a multiple of 24 which in turn means you get an integer number of samples per frame. Makes syncing video and audio much easier.
@@vincentrobinette1507 'crisp' is a subjective thing in human ears. and human senses are not that good. the human mind doesn't work like we think it does. EQ on every stereo are there because human hearing is really not that great lol. the human senses, barely work. if you've seen an optical illusion, you know that.
@@GungaLaGungaI dunno, a handful of fleshy tubes that can detect oscillations of molecules on the order of nanometres with wavelengths spanning several orders of magnitude sounds pretty good to me.
I remember seeing the original Xiph video almost 10 years ago now, and it fundamentally elucidated digital audio to me. Great to see it being shared beyond its original, as it didn't get nearly enough attention.
The best digital recordings are going to be the ones with the most talented mastering. That will have far more impact than high resolution audio. That said when doing mix downs the high resolution seems to be much easier to achieve that great sound.
Note: 20KHz is the highest frequency a teenager can hear. Enjoy it while you can kids because its not going to last. As you age the limit drops. For most adults the CD rate is not just good enough, its way beyond good enough.
Most teenagers have already blown their 20KHz perception with their listening habits. At least we did in my day. This is more the range of what a 5 year-old can hear.
@@ouwebrood497 It's going to sound like a telephone to them. The noise is all filtered out but so are the high frequencies that they can hear but we can't. If a cat had designed the CD instead of a human the sampling rate would have had to be around 180KHz because they can hear up to 85KHz.
I was an avid amateur audio constructor ever since my teens in the 1950s, always striving for the best sound quality I could afford. It took less than a minute to convert me to digital forever the first time listened to a CD! Thanks for this video.
I remember how the first CDs had a high end scratchy sound. It was because the master was mixed that way because they knew the transfer to vinyl would attenuate the high end. So when they made the first CDs they just grabbed the original master tape and sampled it to digital. And the digital faithfully reproduced that extra high end instead of reducing it like the vinyl processing naturally did. End result was digital got a bad rep. Later they remastered with digital in mind and the next release of the same album was much better. So I ended up taking some of my early CDs and tossing them and replacing with newer versions.
@@kthwkr Ah, the (in)famous RIAA equalisation! That chapter gratefully closed forever, along with tracking weights, lateral compensation, surface noise etc that went with expensive turntables and magnetic cartridges. The current infatuation with vinyl is beyond me.
@@jq4t49f3 Add to that the infatuation with tube amplification. I was there when the conversion from tubes to transistors happened and losing the distortion and bias of tubes was a big step forward.
As an audiophile I’ve always tried not to believe in things I don’t understand. I just experiment, listen to things and whatever my ears agree with is what goes. Many times I have preffered the sound of something cheaper, simpler or “less audiophile”. Or found no difference. The key is to pay minimal attention to what people are saying and just try things. Often they are just repeating things they hear or attempting to fit in with a group.
I am using lossy encoding whilst all the people scream FLAC. The amount of parroted nonsese is piling up to literal fairytales of untrue garbage. I play them my MP3 (!, the lowest denominator amongst AAC, OPUS, VORBIS and the likes) created with Lame 3.100 at V0 and they go like "Sounds actually pretty awesome". I tell them it's an MP3 averaging at 275 kBit/s and immediately they claim that they can hear "bad this" and "bad that", and that sound was "hissy" and how they rather want 800 kBit/s. I make an A/B switch on Audacity "showing" that there is no audible difference - nope nope nope, FLAC is the only way and MP3 is bad. Huh.... I even give them the WAV and all the "hissy" things are present in the WAV as well and not butchered by the encoder. Nah, dude, all bad. These people like to keep ultrasonic content and irrelevant audio information at all cost. . . . . . Placebo effects are a thing, but not only with your ears, but also with your brain. Maybe analog to a kid that fell off of a bike: The knee hurts. Ouch, but fine. Same accident, same pain, but now the he kid sees blood on the knee. Waaaaaaaa!!!! Mama!!
@@ZedekThat's the difference between audiophiles and audio engineers. Audiophiles rely on subjective impressions to judge their equipment, with all the human fallibility and bias that brings. Engineers have test equipment to measure it.
@@Zedek Yeah, back in around 2001 (I remember that year vividly because I did this at a unit I was living in when 9/11 happened), I had heard that LAME 256k CBR was supposedly so good that it was indistinguishable from the lossless source. I was running Debian Potato at the time and thought I'd try it out. I started ripping with cdparanoia my favourite CD's and then converting those rips to LAME 256k CBR, to do some tests. I was doubtful, so as is human nature I was searching for unwanted, audible artefacts. I was shocked that these MP3's sounded so amazingly good and then suddenly I heard it! A warble in one of my favourite songs, which was not musical, seemed out of place and I did not recognise. I went back to the source CD and listened......... and there it was, that warble, on my pristine CD, of one of my favourite songs, that I'd never noticed before. I checked other sources of the same song and what do you know, that warble was there too. I now cannot un-hear that warble when I listen to that song. The ultra critical headspace I was in, while listening, caused me to find what sounded like non-musical artefacts in my favourite music, yet at the same time I was completely unable to find anything detrimental added by the LAME 256k CBR encoding process. So yeah, LAME is a pretty awesome encoder and I understand it has likely improved even more over the past 20+ years and that there are some really awesome and ultra efficient open source codecs now (OPUS rings a bell).
@@ZedekDisparity of samplerate and audible lossy compression are not analogous at all, IMO. There are 100% people out there, including myself, who are not lying to you when they say they can hear mpeg compression. This is a vastly different conversation from the samplerate debate.
My best argument FOR the CD-standard is that it is not something they came up with and then proved to be good enough with math. The CD-standard was based on the proven math and was actually extremely stringent for its time. It's not more complicated than that with good electronics.
@@acoustic61, well, that's true, buuuuut we HAVE empirically proven, as seen in this video, that the math holds for band-limited digital signals, and the same goes for the CD format.
@@TokeBoisen Music is more complex than some demo of a sine wave. I've listened to thousands of hi-res digital transfers and virtually every one sounds better than CD. I think it's easier to get better results with higher sample rates. Maybe because steep filters, which are imperfect, and other forms of processing can be used more sparingly. I see no reason not to use higher sample rares. Digtal storage is dirt cheap.
@@acoustic61 oversampling sounds better if any form of saturation is involved. So you're right that it won't matter for just a sine wave. Dan warrel has a good video on oversampling.
@@acoustic61, the sine wave is just the easiest example to demonstrate the concept. As is explained in the longer video from Monty, any complex waveform of a band-limited signal can be perfectly captured and replicated. If it deviates from the original waveform it MUST contain information above Nyquist and is therefore no longer a possible solution. It HAS been documented and verified that phase-differences can impact the ability to capture perfectly if the chosen sampling-rate is exactly twice what you'd want to capture, but that is inconsequential for either CD or modern digital formats where the sampling-rate is much higher than twice the upper limit of human hearing. Additionally, if you ever look at FFTs of hi-res transfers you'll most likely see that there is either no information above 20 kHz, or what is there is just noise. At best that means they just waste bandwidth, at worst it introduces IMD in the audible range. Any differences you are hearing are more likely to be due to a difference in mastering, an increase in gain, or just simply psychoacoustics.
Thanks for mentioning at the end that 24bit and higher sample rate is important for production. Indeed, for recording one doesn't need more than 44.1 or 48Khz, however in production when there are tens of filters and effects applied to a sound, distortion and noise is amplified if the resolution is insufficient.
I can't think of anything that would create either noise nor distortion if the sample rate at which the audio is recorded at 44.1k. Oversampling for non-linear processing exists and for linear processing it's not necessary (and for things like convolution a higher sample frequency makes everything eat up computational resources. My DAW supports 16xOS, applicable to either plugin chains or single plugins, which is plenty. No need to waste CPU power and thus performance on using higher sample rates for things that don't benefit fromthem...
@@simongunkel7457 They produce ultrasonic harmonics, it you don't have space enough on your bitrate to accomodate them, they "bounce" back to the sonic range and they can be heard as "artifacts". So you need space where they can propagate, and can be cut clean when converting back to 16 bits, because all the "trash" is out there beyond the 16 bits resoultion. They will explain this on the next video for sure.
@@framegrace1You are confusing sampling rate with quantization (bit depth). The bounce back happens if you have non linear processing (like all distortion) and the sampling rate is too low. Quantization only affects the noise floor.
@@framegrace1 You are confusing bit depth and sampling rate here. There are good reasons to record at 24 bit depth, but that doesn't have to do with ultrasonic harmonics, but it allows for more headroom when recording and more tracks to be mixed (you lose about a bit for each quadrupling of tracks, so if your project has 64 tracks that will mean your output will have lost 3 bits compared to the recording. And you also lose a bit for each 6dB of headroom while tracking. When 16 bit was state of the art, you had to track really hot, which made clipping likely. With 24 bit, you don't have to track hot at all and won't clip and still have plenty of bit depth to allow you to go ham with multitracking). Aliasing, i.e. ultrasonic harmonics bouncing back from Nyquist has to do with the sampling rate. But as I mentioned you can use oversampling to deal with that. So my recorded audio at 48k goes to a non-linear plugin. It will alias. I enable 16x oversampling for that plugin, which will make the plugin see a 768k signal where the ultrasonics get to live up to 384k and then have another 360k before they would bounce below the 24k where they could end up in my 48k signal. That is then filtered out and the signal is converted back to 48k. But if I send my 48k signal to a linear plugin, it won't produce any harmonics and thus I don't need oversampling and will just run the 48k signal. Convolution is usually the most intense processing and it is linear, but the computational load scales with the sample rate. Note that my line of thinking here requires a DAW or plugins that oversample. I'm using reaper and per plugin oversampling has only been a feature for less than a year and not every non-linear plugin has internal oversampling (though plenty do). But even then tracking at higher sample rates rather than processing at higher sample rates didn't make much sense and now you can get granular with that. If you hit everything with 96k, you will slow down things that don't produce aliasing anyway and things that do will still have bigger issues than if they specifically got 768k signals to work with.
Exactly, processing in the digital domain will degrade the signal, unless you start out with a higher quality already. Just a simple +3dB filter already throws out one bit of dynamic range, so you better start out with some headroom. Somehow audiophiles got wind of the fact that these audio formats exist and that it would be better to listen to this original mastering instead of a downsampled copy, which obviously makes zero difference, unless you can upgrade your ears to higher dynamic range and higher frequencies. Although I'd still say that for typical audio mastering, only the bit depth will actually matter, unless you have some serious slowing down planned. Otherwise the high frequency components would remain outside the audible range either way. Maybe it might protect better against some quantization artifacts when filtering.
I worked for BBCTV for 27 years as a sound operator and was there when digital recording reared its head. This video is most interesting and actually rather reassuring. Thank you!
One of my favorite BBC R&D inventions is sounds-in-sync that used the same concept later used for digital recording with PCM converters to videotape, by digitizing analog sound and then converting that binary signal into a black-and-white television signal that could be sent along the standard black-and-white signal. The BBC used this also to feed FM transmitters where the digital sound would only be converted back in to analog and modulated onto FM in the transmitter. Basically Britain had the pre-cursor to DAB radio running from 1970 onwards - pretty impressive.
@@AudioUniversity TH-cam does not allow a video of someone doing a finger stick test to check their blood sugar, the drop of blood might offend or scare someone...sigh.......
As a member of the general public without any background in sound engineering who always heard of the stair-step wave thing and how digital sound is "worse" than analog because of it... Your explanation is SUPER on point. I was able to fully understand the concept. I give you kudos for being able to create content valueble for specialists and laymen alike. You're awesome! :)
Me too! But I had trouble understanding some points. For instance, I've never heard of the stair-step at all, so I had no idea what it even was, honestly.
@@toomanyhobbies2011 well, he presented a long video with a lot of sound, reasonable argumenting for his point. You'll need more than "he's wrong" to be taken seriously. So please elaborate.
The people that tout "worse than analog" usually compare the most expensive analog solutions to the cheapest digital ones, in which case you're just comparing the relative quality of the equipment. But similar goes for the other side of the argument, if you compare a high-end digital solution to a bargain bin analog one the more expensive one will objectively be better, but again not because it's digital. If you get something decent and keep the signal properly intact from start to finish both will be as good as the other, with some minor changes in the tuning depending on what all is in the chain at what point. I have both and enjoy using both depending on what I want to listen to at any given point. Analog is a bit more prone to outside interference like ground loops or cables picking up noises from a nearby power line, which attracts purists that want to feel better about themselves using a harder medium, but digital is much more convenient and thus easier accessible to everyone. Good music should be able to be enjoyed by everyone, not just elitists.
Thank you for uploading this! I saw Monty's video over 10 years ago, but just a few years ago I tried to find it again and couldn't. This video came up in my YT feed and something told me it might be what I was looking for.
Higher bit depth though is used in the production process to be able to do things like adding two loud signals or other processing functions that would otherwise add quantization noise (ever see the bands in a dark part of a streamed movie?) or blow past the highest signal of a lower bit depth format. The signal is then requantized before mastering back down to 16 bits. The creators of the CD format really knew what they were doing, and had over a century of digital and signal processing research to call upon to come up with the 16 bit/44.1 kHz format.
Same thing for sample rate: They have a use in production. When you want to do non-linear filtering without the higher harmonics getting aliased back down, or because it helps to avoid filtering artefacts, or so you can slow down a sample without it turning really base-y. But those are only intermediate stages, needed to help the mathematics of transformation work. Once production is done, it's all turned back into something more suited to limited human hearing. 44.1KHz, or sometimes 48KHz.
All true. High sample rates and bit depths are very important for production. But it's a complete waste for distribution. But it's not ALWAYS important for production either. Aliasing is not as common a problem as most people think. Not everyone is pitching and slowing recorded content down. And as for the higher dynamic range - it's very helpful to have that headroom in a recording, but again, is often not an issue. A lot of producers use it as a safety net, even though they rarely need it. I'm more excited about the move to record in 32-bit float. That's not about dynamic range, but about never clipping unless the mic itself is overloaded, and never having to worry about gain until you are mixing. It's like the audio equivalent of shooting photos in raw.
Fabulous. I have understood this although at less technical level until now. That testing method he shows is absolutely brilliant and his explanation is spot-on. Those first engineers at Phillips and Sony who created this standard and the Red Book, they knew what they were doing and set the bar at a very high level. This is a fantastic addition not only to your content library but for the general public.
Thanks, @Grand Rapids57! I agree - the bar has been set incredibly high. Unlike developments in video recreation, audio has really stood the test of time since the CD.
No wonder - although the result was wonderful ! Audiophiles were much more inclined to testing equipments… And, what’s more, the human ear is an outstanding tool, much better at discriminating than the eye : visual illusions are plenty, but aural illusions are just a few. And if you want to know the material of a wall, your eyes will easily mess up. Knock it with you fist and you’re ears will tell ! 🤓 Hearing is knowing, as exemplified by the fact that no one will make fun of a blind man, but a semi-deaf man is a funny character because he « doesn’t understand » and keeps making stupid mistakes… In French, « entendement » is what translates as « understanding », suggesting that « entendre » (I.e. to hear) is intimately linked to the knowledge of the outer world. Outstanding video, by the way. Deserves much more views ( sound included), even with the sloppy YT compression !
Philips wanted to stick to 14 bits in the beginning and only a short period before the market introduction the specification was changed to 16 bits under pressure of Sony. The first Philips CD players used 14 bit D/A converters and they implemented oversampling to end up with 16 bit accuracy. Philips already had the design of their DAC’s ready for production so there was no time to create a new design so the clever Philips engineers came up with a 4x oversampling design using the 14 bit DAC to end up with 16 bit resolution.
@@actionjksn they couldn't stick with 14 bits since the CD format was already determined to be 16 bit at that time. Using a 14 bit converter without oversampling would have caused degredation. In the end it didn't cause any downsides in terms of sound quality. Philips was also on top of there game in that period, they produced some of the best DAC's in that period.
Thank you for this amazing video. As an old film sound engineer I used audiotape even without Dolby noise reduction, optical sound, mono stereo, and later digital systems. The most remarkable experience was a CD, a classic concert with orchestra, choir and organ, replayed from a commercial Philips CD player and after that replayed from a similar CD player where all electrolytic capacitors were replaced by other capacitors and copper leads and wires were replaced by carbon leads. The recording sounded much better and you had the idea that you were in the recording place yourself. The engineer who constructed this CD player was clashing with his bosses and with Philips to keep this invention quiet.
If you want to hear a good example of quantization noise, the Game Boy Advance has digital sound channels that are limited to 8 bits. Accordingly, many games sound like they have tape-like hiss, but only when sound is actually playing.
Key to this is limiting the bandwidth of the sampled signal. With proper low pass filtering, frequencies above Nyquist are eliminated and there will be no aliasing noise. However, oversampling can simplify the input filter complexity and the high frequency content (noise) can then be filtered digitally and downsampled to generate the output digital stream at 44.1K or 48K samples per second.
Internally audio production software uses 32 bit floating point to avoid clipping the 65536 values a 16 bit software synthesizer can generate. The final output is 16 bit though. 32 bit floating point is also faster for cpu calculations. In this video only the end results are matter-of-fact for human hearing.
That's why it's 44.1K rather than 40K - need a bit of extra room because real-world filters are not perfect, you can't have a brick wall. Well, you can with oversampling and digital filters, but that wasn't around back when CD was introduced.
It's actually the output filter that restores the original analog waveform. So the quality and steepness (or rather the lack of steepness, and here come oversampling and dithering into play) of the filter are essential. Maybe it should've been mentioned in the video. Edit: Oh, I see Max Nielsen has already explained this in a much more professional manner. So it is with those who write before they read.
Agreed. A lot of people say that it's not a stair step, but the digital chain goes Dots from the PCM file -> Stair Step at the DAC chip -> Smoothed out using a filter. Source: Texas Instruments, maker of DACs
Many people have not grasped the entire chain and seize upon denouncing the stair step saying it does not exist. It might not exist at the speakers, but deep inside, it does exist. That some people fear the stair step at the speakers is perhaps reasonable in the case of an extremely cheap system that makes no attempt to filter and uses 8 bit, 11 KHz sampling which is barely adequate for voice and the sampling artifacts are clearly audible unless heavily filtered, in which case the heavy filtering is noticeable.
@@thomasmaughan4798 Audio University has given out so much misleading information that has led to online arguments that someone needs to do a debunking video. This video title is very misleading
Some of the reasons to sample higher than 44.1 kHz was to make anti aliasing filter less steep to avoiding aliasing. This filter is in the frontend and analogue, so the filter is cheaper and gives less phase distortion. If the signal is oversampled many times you can take advantages to noise shape the spectrum (place the noise in the frequency range you will filter away).
@SR-ml4dn In the original article, Monty does mention that *processing* at higher sampling and depth is perfectly sensible, the original article (which is linked in the description) was a response to "high definition audio". And the video here is a complement to that, clarifying misconceptions about digital audio & audio waveforms.
Front-end analogue filters are a problem because they can't be sharp enough without introducing distortion and noise. Instead we just sample at, say, quadruple the rate (176kHz) so the analogue aliasing filter can be much less sharp. Then we use a digital filter before down-sampling to 44kHz. Even digital can have trouble with the sharp cutoff between 20kHz of hearing and the 22kHz nyquist limit. Moving the nyquist limit up to 24kHz helps which is why 48kHz sampling is so common in newer standards, including digital video formats.
@@rogerphelps9939most/all silicon based sigma delta converters create idle tones and some cause noise floor modulation. This is why engineers like Bruno Putszeys designed discrete PWM conversion methods.
Came here to say this having a masters degree in vibrations for mechanical systems. Aliasing and side lobe distortion is always a major issue when recreating frequencies from sample data with FFTs and I see no reason why it would different here.
I remember the origins of the CD and people dismissing digital as chopping the sound into little bits, that just has to sound awful! But of course most audiophiles are not that technically educated. Maybe just the ones that design and build the stuff. And their opinions are thus also tainted by wanting to sell you something.
@@thomasmaughan4798 Does one have to be all that wise or smart in order to be technically educated in a topic? (In case there was any ambiguity: I don’t mean this as any kind of dig at people who are technically educated in the topic)
I didn't really think CDs would sound good when I first heard about how they work, but I also didn't think of myself as some kind of superexpert on all topics so I just listened to a couple. I had friends who swore the quality wasn't equal to vinyl but I couldn't hear any difference other than the vinyl was noisier. So... I bought myself a fair number of CDs.
I learned this while studying computer science. It was one of the basics required for the bachelor. The same methods/theories are used for all kind of signal processing like WLAN or cell phone networks.
it's also used in television, radio, basically, if it has a signal, it likely at some point in the signal chain, uses the logic of the Nyquist sampling theorem in it's design.
I would guess they use 24 bit high sample rate in music production for essentially the same reason movie producers and Photoshop artists use significantly higher resolution source material than the final product, because when the final rendering is done, any subtle errors in production will be ultimately diminished to the point they are barely if at all noticeable
Exactly, it's about preserving dynamic range of the finished product. It means you don't need to perfectly hit full scale range when you digitize the audio signal. You want to leave a bit of headroom when you feed any analog signal into an ADC so clipping does not occur. If you record at 16 bits and your resulting signal only uses 15 bits because you left yourself just a little headroom to avoid clipping, you can never get that bit back and your product can not be any better than 15 bits deep, or about 90 dB. If you record at 24 bits and your signal only comes in 20 bits deep because you left some headroom that means you have 120 dB of dynamic range so you can hit your 16 bit/96 dB target. Also, if you process everything at 16 bits then any added noise in the processing chain means your finished product cannot meet the 16 bit/96 dB dynamic range target of a CD. I'm not sure exactly what advantage a sampling rate about 44.1 kHz would give you.
Yeah. I have a little pocket field recorder and I record at 24bit, mostly because then you don't have to care much about recording level - as long as it's there and it's not clipping you just adjust the volume in post
@@Fix_It_Again_Tony I'm not an engineer but a guess about one reason of probably multiple reasons why they might sample at 192k: When a studio records a track, each instrument/voice is recorded separately. Each track has overtones and such that have frequencies above the 22k limit. But when you mix them all back together, you end up with interference/resonance that results in some sound being audible to human hearing. A small, almost imperceptible part of the audio that you would hear if the band was all playing live in the same studio room
@@Sumanitu no, 192k is because nyquist thing, with analog filter you will never remove frequencies > 24khz and that will damage signal bellow 24khz during sampling. So low order filtering, 192k sample, then hi order (almost ideal) digital filtering to get rid of everything above 24khz and then resampling down to 48k
@@tkupilik the "nyquist thing" is literally what I'm talking about, bud. You dont want to lose frequencies above 22khz (1/2 the 44khz sample rate), even though they can't be heard by human ears. Thanks for re-explaining exactly what I was talking about...
I have lots of different music formats: SACD, MP3, FLAC, DVD-Audio, Blu-Ray, and on and on. This video is in line with my observations. The original recording, regardless of the format, is much more important. But I still like high-density formats. I cannot tell the difference between ANY of these formats, EXCEPT when I crank up the volume. I can crank up the volume on my giant speakers maybe 5-10% louder (which I love) without distortion with high density formats (and I have plenty of the same recordings in multiple formats). But I always assumed it had more to do with the way the sound is processed by my specific electronics than by the actual bit rate. It varies by specific systems. This video implies that the noise reduction of higher bits may be the culprit, but given all the electronics issues involved, it's hard to pinpoint. Same with vinyl - I can get some unbelievably sweet sound from certain vinyl that just isn't the same from digital, but only with excellent (but rare) specific records, and only with crazy complicated setup and very specific components I've discovered via long-term trial and error. So again, it seems to have more to do with the specific setup than the actual music format. I love vinyl, but I laugh at the complexity of getting truly great sound with it - so many variables.
Thanks! It seems that we are reaching a point where younger generations are getting educated on these topics, and ruthlessly debunking all the old myths. I'm enjoying every moment of it. The same thing is happening in the world of electric guitar. Shout-outs to Jim Lill.
I've been looking that original video for ages since I saw it years ago. It clears up so many myths, especially he confusion between sample points (infinitely small) and the "fatbits" version audio software tends to show (though nowadays some actually show the equivalent of what the DAC will output, with proper curves).
Arguably, that's bullshit too. Whatever you used to capture the original image DIDN'T have any sort of idealized pinhole-sized pixels - it had a surface area of some shape (actually rectangular most likely), and what it captured was the sum of all photons landing ANYWHERE WITHIN THAT AREA. Representing that as a square IS the sane way of saying "withing this specific area, this is the color we captured, and we have no finer-grained information concerning that area". And that remains true either for digital cameras or rolls of film - only there your "pixels" are the crystal grain of the film, of irregular shape...
Licensed professional engineer with undergrad degrees in engineering, physics, and mathematics and a masters degree in engineering AND a professional music producer here. You just got a subscribe. This was one of the best videos I've seen in awhile... and though I theoretically understood this based on DA design concepts, this video spelled it out with such clarity and reason, that I had to comment. Great work!!
Just get rid of the lollipops. The output of DAC is latched, there are no lollipops and the staircase is real -- just microscopically tiny and passed through a filter to get rid of the 44.1 KHz sampling frequency.
@@mc2engineeringprof " When it's reconstructed to analog, it definitely isn't a stair step." It most certainly is! DAC can ONLY output DISCRETE voltages. You give it a 16 bit code, it spits out the corresponding voltage until it gets a new 16 bit word. Then the voltage jumps immediately (within a few nanoseconds) to the new voltage. "Look at the output on a scope. Show me the stairs steps." You CANNOT see 65536 steps on a scope with barely 8 bit vertical resolution! Noise will exceed the stair steps anyway. Now, if you increase the vertical gain to zoom in on a *portion* of the sine wave, and you probe the output of the DAC itself, you will see staircasing. *That is how it works* Some TH-cam videos compare oscilloscopes on this exact procedure. A digital scope creates staircases on *input* depending on its ADC. That's why you use an analog scope for this sort of thing; a digital scope itself introduces staircasing. But what is the resolution of an analog scope? It is noise limited for one thing; 16 bits is a LOT of depth. A typical CRT for television has 512 lines of vertical resolution, it is impossible to see 65536 stair steps when the phosphor dot is already straddling hundreds of these steps.
@@thomasmaughan4798 you're describing audio DACs from the early 90s maybe, they haven't worked like that the past 25 years though. Modern audio DACs use upsampling and sigma-delta modulation to convert the PCM to very high sample rate (a few MHz) but only a few bits of resolution, so the unfiltered output may not look like the original signal but it also does not look like the kind of stairstep you're describing. More importantly, it's kind of irrelevant what the unfiltered output looks like, the external analog filter is essential to the proper operation of the DAC. The overall result is that it will properly reconstruct, to a high degree of accuracy, the (unique) bandlimited signal that passes through the sample points given to it, provided that this signal is within the passband of the DAC (which is at least 20 kHz when using a 44.1 kHz sample rate). The details of how exactly this is achieved is quite interesting but not particularly important to understanding digital audio as a concept.
*proceeds to change all settings on my audio drivers to 44100 CD and leaving them like that forever* It helps so much with eliminating higher processing needs that might end up in artifacts such as stuttering or crackling of the samples. Thank you! This has literally changed how I use studio equipment.
There is a benefit for using higher sampling rates, but it only applies when making the actual recording (I think this is what is alluded to at the end of the video - looking forward to the next video to see if I'm right!). Although humans can only hear up to 20 kHz, physical sound waves include frequencies higher than that. Those frequencies must be filtered out before sampling or else they will get aliased into the hearable range. But no filter is perfect and frequencies above 20 kHz will pass through the filter, though higher frequencies will be attenuated more than lower. So the solution is to sample at rates much higher than the filter's baseband bandwidth so that only the highly attenuated very high frequencies alias into the hearable range.
That's exactly the point. Ans also why amateur recording and reproduction of high bitrate without modification can actually sound worse than lower bitrates. Those are the ultrasonic harmonics that he talked about.
Well if you record a 20khz sound at 20khz, then every sample will get the wave at the same position, so the computer will assume it's just a straight line. So you need at least double that to make sure you can get both the peaks and troughs of the sound wave.
@@goldenfloof5469 Yes, however Nyquist is a double-edged sword. A 30kHz wave recorded at 20kHz will look like a 10kHz wave; that's why sampling high and then digitally filtering can produce nicer audio.
This is called "oversampling" and it's built into every modern analog-to-digital converter, even the cheap ones. There's no point in recording at a high sample rate and then downsampling; the converter does it for you.
@@allochthon I am an electronics design engineer; this is simply not true, sorry to say. That being said, you may be right with respect to half-decent converters which allow for multiple sample rates to be set. The problem is up-sampling and down-sampling with high fidelity uses non-trivial algorithms to implement in hardware. 48kHz -> 44.1kHz is an ugly conversion, for example, and a surprisingly common maximum for inexpensive electronics (though 96kHz is becoming more common). You can't just average data when down-sampling and get the same results as oversampling and then software processing.
Yes. There are several reasons to use higher bit depth and sample rate for audio production (recording, mixing, and mastering). That’s what the next video is all about. This video is just about audio playback.
@@AudioUniversity yes, for playback 44.1/16 is perfect. My CD collection is still growing and I really enjoy how good they sound considering how affordable a great sounding CD player is nowadays.
24bits does not give you more headroom. Headroom is loudness or volume above Odb(VU) or your normal operating level volume! 24bits give better sample accuracy or sound dynamics so does mean you are sending a better (more accurate signal to your effects)
@@rods6405 Depends on how you transform. You can easily have more headroom with 24bit and use it for expanded dynamic range, this is in fact one of the the primary uses of it before your compress it to 16bit at selected boosts.
0:15... actually, this goes against what the "I paid $2000 for my interconnect" audiofools say. The actual experts, the people who actually know about this stuff, sound engineers, acousticians and the like, completely agree with you.
A HUGE thank you! Excellent presentation! The Philips/Sony engineers that developed the CD format were brilliant! I was in college getting my engineering degree while Philips/Sony was developing the CD format so most of this information is NOT new to me, but what was truly mind blowing back then was storage density. My buddy and I were going through the College of Engineering junk pile and found two MASSIVE hard drives, and having just heard what the CD format was going to be, we started crunching numbers to use those junk hard drives and other instrumentation to kludge up an A/D and make our own CD format files from taped live recordings. Well, the whole effort ran out of steam when we figured out that the MASSIVE hard drives could hold at most eight seconds of music. Eight seconds. Oh, and I'm still using the Philips CD player with TDA1541DACs that I bought in 1988.
Good discussion! I stand by my assertion that the best improvement in my stereo was the cheap CD player I bought back in the early 80s. Gone were the days of clicks and pops of vinal! It is interesting that vinal is making a comeback. I think it's the ritual of the turntable and cleaning that's the reason. I did double blind testing on very high-end equipment from the late 70s and early 80s and the CD always sounded better. The only thing close was a $5000 turntable/cartridge combo that was awesome! Sounded almost as good as a $130 CD player with a DDD disk. My best takeaway is it's the music that's important and not the gear! You're far better off spending your money of good quality recordings than expensive equipment. That said... Spend money on good speakers. Only the microphone (which you generally don't control) and your speakers should color your sound. Tube amps color sound in a nice way (even harmonics) but at the end of the day, it's still distortion! Great for guitar amps! And yes, I'm listening to this on an amp with a tube preamp because "tubes"! They are cool and glow! And have a VU meter! And that's through speakers of my own design. I'll switch back to a cleaner amp soon. But the VU meter is cool!!! Thanks for the video!
Higher sample rates are great if you are a sound designer and wish to do lots of time compression and expansion, particularly if you wish to get fewer digital artifacts. There are also times when you want the digital artifacts from time compression and expansion, so it’s all about that your desired effects are. Knowing when and how to use each audio sample rate and bit depth is a skill you get from experience. Great breakdown on the technical side for those not in audio production. Can’t wait to hear your thoughts on how to use each sample rate and bit depth when it comes to professional audio production. I wonder if you will get into more than just music production, but also pre and post production for TV/Film/Radio/Podcasts, Sound Design, Foley, Voice overs, and Voice Acting to name a few… Mixing down to 44.1khz 16bit is perfectly fine and sometimes preferable, it’s only when you start messing with audio’s time and pitch do you really start to see immense utility in higher sample rates like 196. It’s great as a Sound Designer when you get to choose between different sample rates and bit depths, it adds so much flexibility when tweaking sound effects and voices. It can save lots of time and money if you have the ability to tweak sound effects and ADR VO to work and not have to record new sound effects, or especially bringing back in an expensive Voice Actor just because you are limited because of sample rate just how much you can manipulate a particular voice/sound effect… Just my two cents…
Off course high sample rates matter when it comes to sound manipulation in samplers/DAWS/vst-i:s. But his video was only about sample playback for a fixed song with a wave that isn't manipulated. Even back in the 85 when the Fairlight CMI 3 was released, they knew that playing back 100 KhZ samples pitched down an octave would get you really cool sounding low tones that still retained "full quality" with less artifacts! Oversampling is also used in many softsynths when they are doing "internal calculations".
So much disinfo here I don’t know where to start. Without realizing it, you are propagating the very silliness this entire video was made to debunk. I suggest you study dither until you actually understand it for starters. Then do some more reading to find out how top engineers actually do their digital recording. Sound Design is a somewhat low demand version of music, not the opposite. I’m happy for you to feel special about your chosen line of work, but it would behoove you to know what you are saying before you write these ridiculous screeds.
It is a pity that the embedded video did not explain why the step function is not the end-result of digital analog conversion: namely that the signal is subject to subsequent thorough low-pass filtering. Without this information the nice harmonic output of a DAC is surely quite mysterious.
Yes, the missing reconstruction filter has been mentioned in the comments a few time. That it is there is indisputable. That it is not mentioned in the video in my opinion is most likely due to the video being a digital audio marketing video from back in the day. Even the essential anti aliasing filter on the input is not mentioned probably for the same reason, to prevent the viewer from being horrified at having "filters" in the signal path. Other than these simplifications it is a remarkably good demonstration and reflects the real life experience of digitised audio.
Nyquist shows that the stairstep must be converted to Dirac delta conversion to simulate discrete time before lowpass filtering, so it's not the filter, it's the discrete-time approximation.
@@tomgroover1839 Huh? Anything can be filtered. The stair step is the default output of a DAC. Other digital filters may be used and these days often are but a low pass filter will remove higher frequency harmonic noise due to quantisation irrespective of what it looks like in the time or frequency domain. The point is that the lollipops are a representation for the mathematics on paper and are not used electrically. The electrical representation is the stair step until filtering takes place which typically is used.
Apparently some people won't accept the Nyquist result no matter how much evidence you show them.. It seems to be almost a religious conviction. The same people probably have very little ides of what a Fourier transform is. I guess the only way to (maybe) convince them is to put them through a double-blind test situation in which they have to prove that their 'magic ears' can discern the difference. This has been done numerous times, I'm sure but I don't have any cites to hand.... do you know of any good links?
Glad you mentioned the issue with trying to play output that contains ultrasonic content may cause distortion. Many speakers will not have the frequency response to reproduce signals with ultrasound components as they were not designed to and they are very far from the frequencies they were optimised to reproduce best. This will usually be 20Hz-20kHz for a single driver speaker. The reason why no speaker can reproduce all frequencies equally well is that the strength of the electromagnet has to be tuned with the mass of the cone, the range of travel of the cone and the desired frequency response in mind. Play frequencies too low through it and the cone may move too far and physically clip out, play frequencies too high through it and the cone can't really respond quick enough for the smaller fluctuations in the signal producing noise in the mechanism that way. This is also why high end speakers often have multiple drivers fed through band pass filters, the big heavy driver that can push lots of air for that thumping bass is really not suited to producing tones in the higher range that need a more nimble cone or more powerful driver. The problem is putting in a driver that can move that monster through it's full range 15000 times a second and it will hit the limits of it's range of motion part way through the cycle with a 20Hz wave of the same amplitude. On the other end optimising it for a few hundred Hz means the cone doesn't have time to move far before the signal switches at high frequencies. So you can't have your cake and eat it and trying to demand a speaker to do everything will make your listening experience lousy.
Fine work sir! My argument for all the vinyl lovers has always been db range. The maximum for LP was around 60db, not including noise floor. Then add in the pops and crackles after undergoing the RIAA standard. CD has been a great playback and preservation method (so far) compared to vinyl or tape.
playback yes, preservation not so much.. i used to work for an archives compagny and we prioritize CD over Vinyl and tapes, as the support is way durable if conserve in proper conditions. Average life time for a CD to start degrading is around 10 years if i remember correctly. It is way more for others medium
@@mambocountach I've been collecting CD's since the 80's and none of them have failed whether stored in a cool place or warm storage facility. Tape and vinyl don't do well in heat. I say this after retiring from the record industry after 30 plus years.
@@mambocountach I recall there were some CDs manufactured in the UK in the '80s that began to delaminate after 10 or so years. Here in Australia a lot of early CDs were pressed by Disctronics in Melbourne, and to my knowledge none of those discs in my collection have degraded.
@@mambocountach My thinking is that there is no lifespan limit if the disc doesn't delaminate or get fungus(?) between the layers, a very rare phenomena. I have 40 year old Hi8 tapes that play like new, despite the experts claiming a much, much shorter lifespan. Supposedly, stored tightly wound the magnetic fields from one layer will imprint through the substrate to the next, mixing things up. Yeah, right.
I'm very happy to see people start to explain this to the masses. For 2 decades I've been telling audiophiles (extremists?) that they rarely have less than 2-3 bits of noise on the analog part of a 16-bit sample and they don't believe me. Even explaining that 16 bits means 16 microvolts of resolution per volt and that it's easy to get more due to RF radiation around. Not to mention that some such people nowadays use class-D amplifiers and find them good while these ones generally provide less than 8 bits of resolution due to using an insufficient sampling frequency! Thanks for this video, really!
Great video! I want to clarify something about the Nyquist rate: it’s not necessarily that you need to sample at twice the highest frequency, but twice the bandwidth. If your signal of interest is, for instance, between 800 and 1200 Hz, then you can apply a bandpass filter that accepts that range and rejects what’s outside of it. With such a band-limited signal, you don’t need to sample at 2400 Hz, but 800 Hz! Essentially this technique exploits aliasing to reproduce high-frequency signals. Source: I work with RF professionally
Great post! I wondered when someone would correct the faulty definition of the Nyquist theorem. I have also sub-sampled many signals in my days. Also, in the video it was stated that the stair step view is never there. You do get a stair step when you have no DAC anti-aliasing filter.
This is not entirely correct. By this logic a signal between 1GHz and 1GHz+400Hz would only need to be sampled at 800Hz. You're missing the downcoversion step -- mixing with the carrier and low-passing.
@@dmitryjoy Actually you can, without down sampling, as long as the input stage in the ADC has a sufficiently large full power bandwidth. Now you have to bandpass it to only let through your band of interest first though. Sampling at 800Hz with a 400Hz band of interest would be an impossible filter to make since it has to be a "brick" filter though. You also have to be smart about what sampling frequency that you pick. I have done sub-sampling implementations in a professional setting, with RF signals. The linked video explains it. th-cam.com/video/ryJPVHrj0rE/w-d-xo.html
So true, however with CD Audio the dead bands below 20Hz is not worth mentioning as a credible saving. In early A-Law and u-Law phone service codecs they used 8ksample with input filters of 200-2800Hz giving just 2600Hz bandwidth that in theory could have been achieved with 5.2ksamples with perfect fairy dust input filters but simply not worth the trouble for the savings. The input circuitry on VERY high bandwidth oscilloscopes makes use of similar techniques elevating them to near magical levels with parallel sampling of delayed signals as required to compensate for the inadequate sampling rate but still maintaining the input bandwidth. So a 2 GHz input might use 4 x 1GSample ADCs or 8 x 500MSample ADCs plus a lot of DSP magic.
This was a great video. I am a retired electrical engineer. Most of my career was concerned with DSP (digital signal processing) of one form or another. I think that the reason that digital audio mystifies people is that you must have an understanding of some fairly advanced mathematics to understand why it works. Topics such as Nyquist sampling theory, anti-aliasing filters, spectral images, and Fourier transforms come to mind. Most people attempt to apply their common-sense analog thinking to it but that just doesn't work.
What a great video! I was aware of the technical inaccuracy of the stair-step, but I'd never seen someone show the analog signal converted back through an oscilloscope to show it that way. Very interesting. Thank you so much for making this educational material available under Creative Commons, by the way.
I still like the sound of a good well preserved record. its the coloration which seems to have that ear tickling effect, not because its more accurate.
Vinyls still sound great, but it's more because of how they are mastered different from CD than anything to do with it being analogue. Can't have a vinal with so much range that it jumps your needle off the groove and all that.
@@lorestraat8920 Part of what makes valves so appealing to the ears is the harmonic distortion they produce. Its a pleasant effect. And vinyl has a similar effect provided the mastering is done well. But where i truly thought where the debate would hold water, was when recording live instraments to my Teac Simulsync with ampex 456 on 15IPS. Vs recording them to a really good ADC into a computer. The ADC i have is Universal Audio equipped with all Burr Brown opamps and other quality parts. About as good as ADC can get. Using the same mic and mic pre, the Teac sounded better to my ears every time. Its like it preserved a certain warmth and depth that the UA ADC couldn't, and I even tried other recording interfaces as well. Now having the UA ADC in a 64 bit floating point and high sampling did manage to match the Teac in that respect... But Otherwise I am unsure why the analog equipment was seeming to preserve the sound in a more pleasing way. Its why I have tended to keep my recording in the analog domain while it makes sense to do so. Provided it is preserved without degradation from generational copying.
If CD really was as good as these people say then it would have killed off vinyl long ago. The problem for CD is that most musical instruments emit ultrasonic frequencies above 20khz. The CD just cannot reproduce those frequencies, but vinyl can. Also, these frequencies have been shown to have a positive effect on humans. But they don't want to talk about that because they are not particularly smart engineers. But ask the engineers who work for Sony and you will get a different perspective because they actually use their brains.
I became interested in electronics when I was a kid during the mid 1970s. I discovered high fidelity audio as a pre-teen during the late 1970s. At one time, my dream job was to become an EE and work for McIntosh. Then during the early '80s, I added computers to my list of passions. In college, I earned a degree in computer engineering, but I also took all of the electrical engineering coursework as well, which covered things like analog circuit design and analog/digital signal analysis/processing. My professional career has been in computer networking, operating systems, and embedded software development. But I never lost my love for (vintage) analog hifi. (It just took a bit of a hiatus when I was busy raising a family. Now it's back in full swing.) From my educational background, I always understood that the "digital stair step" was a myth, but this video helped further elucidate that fact. Nice job.
There are advantages to 16 bit 192khz sample rates, That is in music production. This is because If you want to time distort an audio track or have it slow down (vinyl effects) It will sound better when it's running at a quarter of the speed because it won't sound so compressed and low frequency. Because all of the ultrasonic frequencies are also being brought down to within audible range so, of course after you do all of your editing, the final production can be 44.1 of course, but for production reasons, it's the equivalent of using vector or extremely extremely high resolution logos, because depending on where you move it and scale it and make it really big or small, it might affect it more or less.
How much of what you do has a dozen or so audio sources? I'm just wondering. I was part of a group called the suspicious cheeselords and there were about a dozen of us. Recording myself I could never tell 44100 from 96000, but with all of us I could pass the double blind. I was wondering, do I have a bad DAC (behringer 1820) or am I actually losing directional data at 44100 through low detail making many overtones/different phases too hard to separate?
Haha! I was about to comment that higher sample rates and bits are very important in music / audio production, but you finally told this at the very end of your video. Great video btw!
The benefit of high bit depth i see is that you can record a quite sound without using the full amplitude range of the recording and after amplifying it's still capable due to the S/N ratio. If you had 16bit but had a quite sound using only 8bits you wouldn't do much with it as amplifying would increase the noise also. Having 24bits can be usefull when you dont use the full amplitude. You can crank the volume up and still get a decent sound with the S/N ratio low without interfering with the music (speech, any audio indeed).
What knowledge this boy has 😮 and the fastest growing channel of audio community. I started following him when he had 77k subs now he is already on 200k+ I learn so much from him. Thank you for sharing! ❤
@@ClosestNearUtopia where in my comment do you see a flex? I'm expressing my appreciation for a good explanation, sounds like you're the one trying to flex here 🤨
This, and one other equation, was the only thing in my four-year electrical engineering degree I'd never heard of anywhere else. Thanks for making a video of it. When my professor proved in calculus that a perfect reproduction of a continuous wave exists in a jagged-looking digital stream, I gasped "what!" and the rest of my class looked disappointingly unmoved by the revelation.
How does this work? What if instead of fifty samples clearly visually suggesting a sine wave, I just have two samples? Is there somehow a way to reconstruct the original even then?? I don't know calculus so I know this may be a hard question to answer.
@@krisrhodes5180 Trying not to be too jargoney and also hoping I remember it correctly... "the Fourier transform of a rectangular pulse" can be applied to the concept of digital sample playback if each "stairstep" is transformed individually... the sinc functions overlap and the resulting waveform looks more and more like the original as you add all the harmonics of the stair steps. He used ancient overhead transparencies to demonstrate it but I'm sure it could be done in Excel or MATLAB.
Fun fact: PVC molecules are big, and vinyl grooves are small. Together, these result in an equivalent dynamic resolution of 12 bit. Under ideal laboratory conditions. But yeah, vinyl is "analog" as opposed to "digital"... that's just words and concepts. In physical reality, every 16 bit delta-sigma converter from the mid 90s is superior by orders of magnitude.
There are vinyls I like. And universally they are of medium to large groups, not soloists. CD is just better when it comes to signal to noise. I just wish that CD were higher sample rate for the orchestras.
It’s the AD that matters. Higher bit depth and sample rate, would be more important on that end, during capture of the sound. There is rounding and/or truncating happening there. It might sound better at lower, but it is a closer capture of the signal at higher rates.
Thanks for taking Monty´s great video to the "next generation". Used the original a lot in hifi-forums, not sure it helped, because many audiophiles share flatearther-vibes. But still one of the best explanations.
If I mixed in 16bits, no one complained or even figured it out. I bet no one cares if the track is cool and has sufficient "musical nutritional value" in it xD People seem to care more about the content of the music than the quality or bit depth. Even if a top track is released on old tape, people celebrate it more than a track that doesn't touch them and that in 24bit. At least that's how I feel xD
That's a great point, 4N4LOG! The most important element is the music itself. However, there are some reasons why you probably should still use 24-bit audio for recording and mixing (when possible). It's not a deal-breaker, but it definitely helps during the recording and mixing phase! Playback should be 16-bit.
I remember reading that even the best audio systems in the world have the noise that equals merely 18 bits one(so no idea why 24 bits exists at all). And unfortunately there are people who just don't believe Nyquist-Shannon sampling theorem no matter what, they somehow hear the difference...
Thank you, I now understand what I don’t like about my nephew’s band’s music. It’s as I thought, the sample frequency is to low. They’ve been compensating by over processing the signals.
A long time ago I wrote a program to reduce the bit depth and sample rate of audio recordings so my students could hear what difference it made. One of them had just bought a CD of Black Sabbath's Greatest Hits so we gave it a try with my program. However, Ozzy Osbourne's vocals sounded exactly the same unless the bit depth was reduced to 4...
When I saw the thumbnail, I was like "yup, that's what the LPF slightly below the Nyquist frequency is for, don't people know?" And then it hit me. No, people don't know. Great video, thanks!
The audio quality is quite sufficient for the CD format, it reminds me when I went from Vinyl to CD I was especially impressed by the dynamics and no background noise... The high fidelity is there remains to choose good speakers that will define the timbre of what we listen...
The problem with this simplistic experiment using a single frequency, continuous sine wave is that it completely eliminates any discussion of phase shift. Go to a high end audio store and listen to some music recorded on a CD that is played back on a player with 4 and 8x oversampling. The difference will astound you. Why? Oversampling permits the use of a lower order filter to reconstruct the analog waveform from rhe samples. What you don't hear discussed in the video is that filters (that reconstuct the signal) have varying phase shifts at different frequencies. Phase shift is time delay and that in turn makes the tones you hear move spatially in the audio you hear. The closer to Nyquist the sampling rate the sharper the filter rolloff needed for reconstruction which in turn necessitates a higher order filter with, you guessed it, a much more distorted phase response. So while 44 kHz audio meets the Nyquist criterion to recover the frequency content, it won't be sufficient to get all the phase info you need for crisp sounding stereo audio. Oversampling is critical for good audio. Don't believe me? Like i said, go to a high end audio store and hear for yourself!
Oversampling in the DAC is digital interpolation between the recorded samples. It doesn't add any information. It just means you're doing some of the filtering digitally and the rest analog.
12:23 if you know how loud a 90dB sound is, a 120dB sound is 1000x louder than it (as the decibel system is an exponential system, +10dB is equivalent to 10x the volume)
The dynamic range of vinyl was at best about 70 dB and CD is 96 dB or at the most 120 dB with dither, but yes, the last 40 years has been ruined by the "Loudness Wars". So sad. I haven't bought any music in the last 20 years.
The same could be said back in the day of analog recording. Hi-fidelity is usually saved for classical, with most pop and rock being recorded with compression play-back in mind, ie, AM/FM radio.
@@garymiles484 you're missing out, tons of less well known artists are producing beautifully mixed and mastered music. tired of hearing old people complain about music these days lol you all just aren't looking and listening hard enough
@@SublimeSynth I am an old fart who started listening to music in the late 60's. Just because the recording process is great does not mean the music is great. Lucky for you to find new music you like. I had given up on new music until I came across rock from Japan!
The simplest way to bust a stair step myth is to discover some of the maths included in signals. With a system that has just enough bandwidth to play 20kHz signals (analog low pass at 22kHz or so) you are just not able to produce this kind of stairs because replicating those stairs would require huge bandwidth way beyond the signal played. So when there are samples every 44100th of a second, the "stairs" simply are not going through the low pass that should always be there after the conversion.
Exactly, the sad part is that mention of the filters was left out of the original video, I think for pro-digital propaganda reasons. Without the anti-aliasing input filter and a extraordinary microphone a strong 30kHz signal (consumer ultrasonic cleaners are often in the 25-37kHz rage) would alias to 44.1-30=14.1kHz and create significant distortion in the digital steam from a noise source that no human ear could hear. Without the output/reconstruction filter the 44.1kHz square wave component of the very real stair step wave would result in lots of wasted power in the output amplifier stages and perhaps overdrive any ultrasonic capable tweeter units but still remain inaudible to humans unless it caused clipping or other non-linear distortion.
@@KallePihlajasaari It was left out because it's an implementation detail of how a DAC works, which he didn't discuss because it isn't important for the topic of the video. What matters is that you can reconstruct a bandlimited audio signal from its discrete samples, provided the sample rate is more than twice the bandwidth, and a modern audio DAC is able to perform this reconstruction with excellent fidelity. How this works is interesting too, but not needed for a conceptual understanding of digital audio and its properties.
@@MatthijsvanDuin The lollipop representation is even less needed for any understanding because it is simply a graphical representation of the digital data stream and it NEVER appear anywhere but on paper and never in the output and not even really anywhere in the real world unless someone was looking at the differentiated output of a radiation detector, which is avoided because pulse height is usually integrated so it can be measured with more accuracy and ease. Look at the clickbait thumbnail of the video, this is propaganda and it is a crying shame someone has to repeat/reinforce it to pander to new generation of confused digital audio acolytes. Digital audio does what it does because of mathematics and engineering, not because of smoke and mirrors.
This is one of those videos I hope I come back to in ten years and completely understand. Like, I know there's some wildly smart and mind-blowing stuff being talked about and shown, but most of the video goes over my head. Still, it's fascinating to watch. Thank you so much for putting this content out for people to watch! I can't wait to fully understand this!
Thank you so much! This was extremely informative. I've always wondered about this, since I couldn't hear the difference between "CD" quality vs higher bit and sample rates. I also enjoy all of your other videos! Kind regards!
RIGHT ON! 😎👍👍 Being old af, we learned all this when digital audio was first hitting the market. It’s kind of a shame that you still have to walk people through this. I think people are confused by the higher data rate that computers often need to PROCESS audio. This is a completely unrelated function.
Music is not a single sine wave of invariable pitch and volume. Thus you were fooled by this video. Sorry. In addition, bits don't equate to dB. Bits simply define a resolution range. You can use 16 bit to define 65536 different volume levels between 0dB and 1dB, or 65536 different volume levels between 0 and 1000dB. So here too you were fooled.
Good and informative, (old CS/EE engineer with audio interest here). Remember back at uni in the 90s we had labs doing exactly this, learning about different types of wave reconstruction from digital data in DA-converters (there is more to it than the video mentions but the conclusion is very valid). One thing that is not mentioned with dynamic range is that there has been historically hard to have a standardized level for where to put the mean audio level in recording digitally. This is what is called the "loudness wars" where listeners tend to like music etc at higher volume. Also the reason commercials are usually at a way louder volume. This drove recording industry to record at higher and higher levels. So a lot of CDs actually had a very high average volume meaning that it was less dynamic range to use. This made them sound inferior to what the CD-standard actually allowed. A lot of CDs were "badly" recorded so to speak. Ideally you would want the digital recording to be fit within the dynamic range in a way that maximizes the format. The effect of reduced dynamic range is similar to how "nightmode" sounds. This is one reason the classic vinylrecords can sound better than CDs. The format in itself is wildy inferior but they were mixed a lot better. There is also another issue related to this and that is the intentional compression of dynamic range to sound better on devices that are not able to accurately reproduce sound. There are some very interesting soundclips comparing speech and music with and without artificial dynamic compression. This is a very big area with lots of opionions (some mixers feel the compression is part of the style). Personally I want my audio to have as much dynamic range as possible. To me it sounds a lot better, subjectively there is more texture in the music.
@@jim9930 Cool setup :) Yeah recording technique makes a huge difference. But with compression I’d did not mean compression as in bit/bandwidth reduction but as in compression of dynamic range. You hear it is a lot of modern recordings. If someone is singing it is the same volume regardless if they are belting or if they are whispering (a bit exaggerated but not too much). This is done in software usually for aesthetic reasons and for it it sound better on poor devices (cellphones, laptop, iPad etc etc). Have listened to some horn systems and they have immense directivity and effectiveness. Tend to get very very big if you want to do good bass with them. Personally I have tinkered a bit with something called orthoaccoustic speakers that utilise room reflections more to create a more enveloping soundscape.
Watched Monty's video right at the beginning of my journey into digital audio and video. Really solid and great deterrent for a lot of the audiophile nonsense lol
Thank you so much for making a video about this topic. It was high time, that someone with the proper reach does a video and debunks these myths. We really need videos like this! It’s just a shame, that nobody of the reviewers, who keep „selling“ a high sampling rate or bit depth on audio interfaces, will never really care to watch this video.
Problem is they CREATED far more false myths than they debunked. The only "debunked myth" was using the stair-step as an [over]simplified way to visualize how limited sampling rates always lose data. The actual mathematics for how sampling loses data is more complex. It comes down to using mathematical guesstimates in the reconstruction filters during the D-to-A playback. But didn't they just disprove that? No. Real music isn't the simplest wave theoretically possible--the 1kHz sine wave. Absolutely No. Real music is a true ocean of multiple waves and overtones with ever-changing apex and trough points which are escaping the sampling rate and have to be "guessed" by the reconstruction filter algorithms. These algorithms create all manner of "false local minima" and "false local maxima" -- basically you might call them hallucinated soundwaves artificially interjected into the playback. Reconstruction algorithms therefore introduce false and lossy artifacts during the D-to-A playback. What Nyquist can do for a single invariable sine wave is true science, but only an idiot mistakes that for variable waves some of which aren't sinusoidal. Moronic and utterly sophomoric and just-plain-dumb-as-f\/ck is the ignorant ineptitude of using a 1kHz sine wave of unchanging volume, as a proxy example for recorded music. Real music presents mind-numbingly difficult mathematical and engineering challenges which lower the quality of all digital playback at ANY AND ALL sample rates currently used by humankind. Want proof? Until you can hear digitally recorded cymbals sound like real metallic rainbow shimmers rippling through a 3D holographic room--which you can't--you have LOSSY LOW FIDELITY. 50 years from now they'll laugh at the arrogance of those saying we can't surpass 16/44k1. At 44k Hz sampling, cymbals sound like some someone pronouncing "ch ch ch" dubbed over the sound of shredding paper.
In 1984 Wendy Carlos released the album digital moonscapes, which was created with digital synthesis, and recorded digitally. Dire Straights 1985 album Brothers in Arms was digitally recorded. There was extensive testing done at that time sending analog audio through digital conversion and back. it was determined that literally no one could tell the difference between the converted signal and the original. In my mind, this has been settled for 40 years.
FINALLY ! Someone besides me finally GETS IT ! This is possibly the best explanation I've ever heard on the topic of digital audio. And I've been doing this long before there was such a thing (62 years and counting.) And yes, 96 dBs of RMS dynamic range is far more than enough for any reasonable playback. I typically mix and master at 80 dBs SPL (at 3 ft.) and that's plenty loud enough. Once again, excellent presentation, thank you ! Bill P.
And here I am at 71 years old and I can barely stand the 25kHz scream of the dentist's scaler. Those things are LOUD if you can hear them. As I age I'm losing my high-frequency hearing, but I can still hear these supposedly quiet machinex. One of my sons has the same range of hearing, and can't have the scalers used on him. Silent Dog Whistle? No such thing as fas as I'm concerned. As for digital music? The problem I experience is the lack of dynamics. MP3 is the worst thing that ever happened to music.
Maybe you just hear the subharmonics when you are at the dentist. As to MP3, read the research by Plenge, Kürer, Wilkins, who came up with the mathematical model of the ear in the 70s, i.e., they analyzed what the human ear can hear and what not. The people at Fraunhofer then took their results and used them to create the MP3 standard...
10:27 That's only on Type 0 tapes. Quality Type 1 tapes definitely don't sound like they have 5-6 bits of S/N ratio, as shown by VWestlife and Cassette Comeback, when the level is adjusted properly. They are closer to 9 bits. And the hiss can also be mitigated quite a bit with those mains powered bulk tape erasers, if your tape deck doesn't erase it properly.
Very interesting, thank you. In my youth I had a hearing range (like you said), of 16Hz to about 19kHz, and an 'awareness' up to 25 kHz. I now have 67 year old ears and a very reduced audible range, so my 30 year-old record/CD/radio setup is quite adequate and still sounds as good as it did when new.
5:29 this is only true if there are as many or fewer waveforms than sample points, which of course will always be true when there's a single sine wave. I'd like to see this experiment replicated with 20 different frequencies.
One effect not explored in these experiments is the phase of the signal. As the pitch of the signal approaches the Nyquist limit, while you still have the frequency itself preserved as a pure tone, its phase will be more and more forcibly synchronized to the samples. This interferes with binaural listening for which we use stereophonic sound reproduction. This would be why it would be that played from a CD, while a violin or a trumpet would be easily localizable to the human pair of ears, a cymbal, full of extremely high frequencies and transients, will tend to sound as if it is coming from all over the place, and not at all as if it comes from where it had in fact been located. True pure analog recording does not have this problem.
I think 400Hz is the frequency where binaural hearing changes from phase to amplitude detection. At 10kHz, for example, the wave length is 3cm, and the ears are too far apart to detect a phase difference, so the brain uses amplitude differences to localize the sound. At least, this is what I seem to remember from Dr, Plenge's course on the psychology of hearing at TU Berlin.
No. Frequency, amplitude, and phase information are all accurately preserved. If the video had used a cymbal crash (limited to 20 kHz) rather than a sinewave, the input and output waveforms would still match (assuming the digital processor box has good filters).
It would be interesting to see what a non-sinusoidal signal turns into with this conversion and how much noise there will be in the final signal. In the above example, an excellent sine wave can be obtained from a sine wave simply because this is how the digital-to-analog converter works.
This is exactly what I was thinking. If by default the DAC is smoothing the waveform assuming a single frequency (clean sine wave) then when you modulate some additional frequencies in there may be a lot more difference between the input & output.
@@MatthiasSchaffSo what? Is sawtooth or square signals not audible? Or no visible on osciloscope? But such signals will definitely show how analog-digital-analog transformation works with all features. Clear sine is not a good example to show that there is no loss with such transformation of signal.
@@MatthiasSchaff the key is using something other than a single frequency. This could easily be established by modulating several frequencies together for a comparison. If a DAC is designed to always smooth as if it is a clean single frequency sine wave then a lot can be lost.
@@GardeningSolutions Exactly. And it so happens music usually is made up of many frequencies all at the same time ! Why this is absent from the discussion here except in this subthread makes the mind boggle. Not to speak of unpitched sounds !
I never even considered the idea that captured samples can only result in one possible output. That revelation is kinda of a game changer for me. Great video!
But that's the point in the video where suddenly there's no explanation where one is needed. :/ How can there only be one possible output? He shows someone drawing multiple possible outputs, then just says "turns out, nope, you can't." Why not? What if there had only been two or three samples? How could there be only one possible output then?
@@krisrhodes5180 if you draw a shape that is clearly wrong but hits the sample points, the you have tried to capture something that has way too high of a frequency for your sampling rate. Thats why sampling frequency needs to be double of the intended frequency range.
Okay I think I may be starting to see how this works. If I have just two samples, then the highest frequency this works for would be one (over the same time as the two samples). I can imagine that if I'm told "these two samples represent a sine wave at a frequency of one or lower," maybe there's a unique solution there. (Still feels like there should be a solution at every possible frequency one or lower but I can believe that feeling is wrong.)
@@krisrhodes5180 It's called the "Nyquist-Shannon sampling theorem", and I think if he went into the math in his video he would just lose the audience.
I said it before, but I am going to repeat it here: these videos of yours are priceless when it comes to bring facts into discussions around audio restitution. So many unscientific arguments are thrown at people's faces in audiophiles forums that debates often end up becoming of an almost religious nature. Pedagogical ressources like yours are therefore highly needed and most welcome. To be frank, though, equipment manufacturers are also to blame for this. Originally, these discussions indeed sprouted from a small portion of the audiophile community who had a grudge against the CD format (for reasons I still fail to understand). Once that debate took off however, many manufacturers of consumer audio equipment started playing into it by feeding people's demand for pointlessly over-sampled signals or obsolete analog devices disguised as "high-end" audio equipment. It all felt very much like how, at about the same time (in the late 90s and forward), computer manufacturers would focus their public communication on the high frequencies of their processors, while integrating those in motherboards with otherwise sluggish buses (making said high frequencies totally inoperative). Now, if I may make a suggestion: You should at some point make a video explaining the technology behind Class D amplifiers. On the topic of debunking myths about "jagged" signals in digital audio, one could hardly find a better counter-example than amplifiers who literally can only spit out ones and zeros on an audio line, while still allowing faithful audio reproduction. Just my two cents.
Thanks for the kind words, Going Modular! I'm glad you find my videos valuable. Also, I'll add Class D amps to the list of topics to cover in future videos. Thank you!
let's also not forget that old "digital" storage media do have degradation unlike modern flash memory or hard drives with all the software making sure the bits are in the right place. early digital formats suffers from data corruption and it ruins the audio result despite how good the decoding part is.
@@fltfathin It was true to a point for a few recordable media like the early DAT format for instance, but that concerned practically few people, and only in the professional range. The dominant consumer digital non-inscriptible format, i.e. the good old CD, was (and is) pretty much flawless (unless mechanically corrupted). On the other side, people do not realise, quite often, how rapidly analog recordings on magnetic tapes used to degrade. Most of the time, unless the storage conditions were optimal and the tapes of high-end quality, signal fading and cross-talking between tracks would become critical after 10-15 years. Not to mention the physical degradation of the tape itself, which sometimes came first, literally making it crumble to dust as it was played back. And of course, not all studio had the space (or the will) to simply store most of their productions. Most pre-mixed multi-tracks recordings are lost (but then, that is also true of digital recordings on physical media _ like the DAT, again _ when Megabytes used to cost good money and take storage space).
Very good video. But... The professor said it -- distinctly and deliberately -- then glossed over it: "Band limited signal." This means that the ONLY waveform the CD is capable of reproducing without distortion at 20kHz is the sine wave. Feed the system a square wave (which theoretically requires an infinite bandwidth), and you get a delightfully pure sine wave out, which is an egregious distortion of the input. Granted, if the human ear is also band limited (can't hear over 20 kHz), then the perceived reproduction is perfect! I'm not complaining about digital (love it), but this fact should always be mentioned in a discussion of the topic.
Very interesting video, but we were only looking from the current standard to higher frequencies and bit depths (I know, there was also 8 bit). What I would find way more interesting would be at which point the signal will get distorted if we lower the sample rate and/or bit depth. Obviously, if you would use only 1 bit, the signal would not be presented very well, but at what point does the distortion begin?
I believe the 44.1 kHz audio used by audio CDs, Was really because at the time with Sony PCM, that was the maximum that could be recorded as data on U-matic videotape for PAL and NTSC standards. And later on, Betamax was often used for digital recording distributions to the CD pressing plants. This was before DAT became the norm in the recording industry.
Ahh penny dropped moment. I took a tour of a CD pressing plant in the 90's when it was all quite new, and we were shown the process from start to finish. When the engineer showed us the source material as a videotape we screwed our faces up, I guess thinking it arrived in analogue form. As an aside, they had to destroy a large proportion of the pressed CD's as they had spots in them, usually black. They played fine, but consumers would think them faulty and return them. Another thing they showed us, and gave us a sample of which I still have today, was the ability to press a hologram into the disk instead of printing the tracklist or whatever. Looks amazing, but I have never seen it used on a CD release.
It's all about context. Which is also skipped over here a little bit. On it's own, yes 16 bit DNR is enough. Unfortunately we don't use a DAC on its own, there will be an amplifier behind it. This amplifier also amplifies the noise. Just in a every day home setting this is often not a big deal. In a very quiet (mixing) environment, speakers that have active filters and very high efficiency high frequency units, the noise can be easily be audibility. The same goes for a big line array system with like 24 compression drivers with a sensitivity of 108dB/2.83V each. How do I know? I develop these systems and I can tell from 15 years of experience that in both cases those noise levels are clearly audible for every human being. There is also a good argument of having some safety margin. Although that's more on the recording site. When music is mastered, there is a lot of EQ'ing going on, you also don't want to clip your signal. Both combined will already give at least a 24-32dB penalty on your DNR. Resulting in an effective SNR of just about 75dB or so. Similar things can happen when it cones down to sample rate. So while I agree with the message that people shouldn't be so worried about DNR/SNR and samplerates as well as debunking some myths, this video still gets a thumbs down from me because it's extremely over simplified. It doesn't explain why some of those higher numbers are actually very useful. Therefore also spreading misinformation.
An amp's job is to amplify all frequencies equally. The db's will stay the same while the dbm's will all increase -- there's no reason why an (excellent) amp should amplify the source noise more than the source signal. (the amp will add its own noise, to be sure). You're totally correct about the safety margins -- that's why we master and digitize at much much higher precision/frequency than the output. And yes, I am yet to see a speaker that can reproduce an even 80db range without distortion 🙂
@@dmitryjoy That's true, the issue is when there no signal present, it will still amplify the noise. Resulting in audible noise at quiet moments for example. Which is just a small annoyance at like monitor speakers, or can be very disturbing for a professional musician with an active floor monitor (with a compression driver) during a concert. But during just regular play at max volume, yes you're correct.
Very impressive content. This is a topic I am well-Informed about, and a misconception that I’ve explained to people on a few occasions. You did a fantastic job of explaining the fundamental fact that the PCM transcription can only be decoded into an analog duplicate of the analog input signal. Noise floor and headroom of the analog components in an ADC and in DAC for playback will matter if you are interfacing with analog audio equipment, so gain-staging and phase and the usual analog caveats still apply- and there absolutely have been very poorly-implemented examples that set bad examples of what digital audio ‘sounds like’- .mp3 stereo 44.1kHz at 128kbps was the norm for a long time, and some .mp3 codecs performed very poorly at properly encoding high frequency content, there is very obvious aliasing that can be heard for everything above 13.5khz. I clicked on this video mainly to check and see if you had your facts straight, and man you sourced an excellent video explanation also. I did not get to this understanding of the PCM wave function on my own studying until I was about 40 years old, after spending about a decade as a telecommunications technician and getting to the point of telecom engineer. I’ve been making music with computers and instruments 30+ years, studying recording technology and techniques, and building my own synthesizers- this level of technical detail is satisfied that you didn’t gloss over any important points and that you drilled down to the absolute truth of the matter without resorting to an analogy (pardon the pun) or a metaphor to explain why. I was on the edge of my seat waiting for you to explain why higher sample rate/Nyquist is desirable in digital audio production, and might be an interesting/informative counterpoint to mention that some production choices intentionally lower the sample rate/bit depth and collapse mono in order to induce digital artifacts during conversion, such as aliasing and quantization loss (the Lo-Fi phenomenon, 12-bit samplers and wavetable synths making a big comeback, the proliferation of bitcrusher FX, etc)
Yes some codecs were dreadful. I remember satellite TV undersampled their transmissions on many programs in early years and sound scores had nasty and painful artifacts. Yamaha's TwinVQ codecs were very good, but all suffered with bad artifacts when the sampling was cut too far.
A couple years ago I re-released an old album of mine re-mastered from original tapes. Lossless 24-bit FLAC is available, but people keep asking me "when are you going to release this on vinyl?" It would sound worse on vinyl, especially since the album is 47 minutes long. It was quiet on the original vinyl with bass turned down by George Horn in mastering to keep needle from skipping. It would probably have to be 2 or 3 EPs, 45 RPM 12 inch to come close to the listener experience of the CD quality (16 bit) digital files.
What concerns me with these experiments is that they are done on pure sine waves. Audio is rarely a pure sine waves. High-frequency (including in-audible) harmonics are additive. So a bit depth helps to re-claim some of these harmonics for a purer sound. It's not necessarily only about the noise floor. It's also about recreating the harmonics on the signal. But it is a good demo none-the-less. That being said, most DAC's nowadays do not do the stair-step method, they climb the voltages between the points so you get more of a "lined" slope. instead of a histogram looking thing.... Modern DAC's are amazing, and represent the MPEG-destroyed digital audio signal to it's very best!
I can think of only one purpose for high sample rates; recording with the intent to slow down or stretch the audio while preserving quality. Does someone more informed than me have another take?
All true, and thanks for an excellent video - it explains it very well! 16-bit/44.1kHz really is enough, despite what some so-called audiophiles otherwise believe (or pretend).
i don't even bother mastering at hi-res because i don't think the audiophile crowd will particularly care about the music i produce, let alone the ones who simply want to enjoy the music (this is where we should be prioritizing about, anyway).
Good point, AI R! Plus, in my research for this video, I found that 192kHz audio will probably have even LOWER fidelity when played back on most systems. The intermodulation of the ultrasonics actually creates audible distortions! So, a true audiophile wouldn't want you to master in high-res formats anyway, I'd say! Read more about that here, if you're interested: people.xiph.org/~xiphmont/demo/neil-young.html
Two arguments: First, we don’t listen to audio tones. Recorded music is nothing close to a sine wave. Live music is represented by complex fundamentals and overtones. That’s how we learn to differentiate between a symbol crash and an automobile crash or between an oboe and a clarinet. Comparing a DAC to analog with pure audio tones is something else entirely. Second, if one grows up never hearing a purely logarithmic recording, it wouldn’t be possible to realize a difference actually exists. Although, digitization brings about affordable,and compact reproduction equipment at an order of magnitude less in cost compared to a comparable analog system, it also comes at a price. Finally setting a music CD as the ultimate goal to analog, Roy Orbison said it best”Only in Dreams” unless pure sine wave tone recovery is the goal. Anything in between introduces coloring. At best, a perfect digitization can only “approach” the complex sound we hear.
I worked for Harman in the mid 90’s & we did blind tests at 44.1, 48 & 96k. Everybody could hear the difference between 44.1 & 48 but most couldn’t hear any difference between 48 & 96. I do all my work in 48k now
Part 2: th-cam.com/video/VSm_7q3Ol04/w-d-xo.html
Watch Monty's full video here: th-cam.com/video/UqiBJbREUgU/w-d-xo.html
Thanks to Monty Montgomery and xiph.org for making this information available with a Creative Commons License!
Some interesting insights into DAC reconstruction filters can be seen in the two application notes by Analogue Devices AN-823 and AN-837 for Direct digital synthesis applications but theory is similar.
A video describes the same details th-cam.com/video/dD9HC1GThZY/w-d-xo.html
The original video featured here left out the reason that the stair step waveform is not visible on the output. It is because of the reconstruction/output filter after the DAC.
Denying the existence of the stair step is a little disingenuous though.
AudioUniversity, huh?? I’ve lost count at this point of the misinfo provided as opinion, and downright incorrect info offered as education. Too bad TH-cam just doesn’t care. But I do. I now know this channel to be a safe space for cr*p, home to clickbait nonsense.
@nicksterj Were you directing the question at me? The glossing over of the fact that the output from the DAC is a stair-step before the reconstruction/output filter which is sometimes digital but traditionally analogue. The lolli-pops only exist on paper, in the digital realm it is just values. Once it is electrical it is a continuous curve (voltage or current) that will be slew rate limited but it will not be discontinuous. The stair-step is the electrical output you expect from an unfiltered DAC, trying to promote that it does not exist is not correct.
The higher db really helps w/hearing corrections & poor audio can't handle those levels :)
ECCO LA VOSTRA SINUSOIDE
Something not mentioned is that the raw DAC is followed by a low-pass filter at the Nyquist frequency. So, even if the stairsteps are there, it's at the sampling frequency and would get filtered out.
Was thinking exactly the same thing. This is correct !
Exactly. Thank you.
And you filtered out the high frequency sound, too. But anyway with powerful headphone all days long can't hear them anymore. a 44kh give you the capacity to produce 22kh square wave sound. To produce decent 22kh sine wave, you need a sample a least 8 time higher. You can use Bezier algo to reshape the sound wage, but it not sound an accurate sound reproduction anymore.
@@omegaman7377 you have that backwards. With sine waves as your basis function it's impossible to perfectly model a square wave without infinite sine waves. The Nyquist frequency is plenty good for sine waves because it IS the basis function of the model. You can apply more advanced techniques to reconstruct beyond Nyquist, but not in a general way, so it's not useful for audio applications. We can do this in image processing (though with some additional information which is kinda cheating)
@@omegaman7377 Square waves have higher order harmonics, technically infinitely high, compared with their fundamental frequency. A pure sine wave is ONLY the fundamental, so you have your sampling idea completely backwards. Sine waves require lower frequency sampling to recreate without any assumption about the input signal, because you only have to reproduce the fundamental.
There's a reason to use 48khz in movies/videos which has to do with the fact that it's a multiple of 24 which in turn means you get an integer number of samples per frame. Makes syncing video and audio much easier.
My experience is, it sounds better, especially the high end. Comparing CD and DVD side-by-side, the DVD is a little more "crisp".
The only correct answer.
@@vincentrobinette1507lol
@@vincentrobinette1507 'crisp' is a subjective thing in human ears. and human senses are not that good. the human mind doesn't work like we think it does. EQ on every stereo are there because human hearing is really not that great lol. the human senses, barely work. if you've seen an optical illusion, you know that.
@@GungaLaGungaI dunno, a handful of fleshy tubes that can detect oscillations of molecules on the order of nanometres with wavelengths spanning several orders of magnitude sounds pretty good to me.
I remember seeing the original Xiph video almost 10 years ago now, and it fundamentally elucidated digital audio to me. Great to see it being shared beyond its original, as it didn't get nearly enough attention.
The best digital recordings are going to be the ones with the most talented mastering. That will have far more impact than high resolution audio. That said when doing mix downs the high resolution seems to be much easier to achieve that great sound.
me, too. Revealing
Note: 20KHz is the highest frequency a teenager can hear. Enjoy it while you can kids because its not going to last. As you age the limit drops. For most adults the CD rate is not just good enough, its way beyond good enough.
Most teenagers have already blown their 20KHz perception with their listening habits. At least we did in my day. This is more the range of what a 5 year-old can hear.
Yeap, can't hear the whine of a CRT monitor anymore. But, boy it bugs the hell out of those youngins to play some Nintendo on one...
Sometimes I wonder what my cats actually hear when I play a CD. (That's a lon time ago to be honest.)
@@ouwebrood497 It's going to sound like a telephone to them. The noise is all filtered out but so are the high frequencies that they can hear but we can't. If a cat had designed the CD instead of a human the sampling rate would have had to be around 180KHz because they can hear up to 85KHz.
nah, im still annoyed at the "rat" repellent, when i go to a certain store that uses it, its painful.
and im way past the teenage years.
I was an avid amateur audio constructor ever since my teens in the 1950s, always striving for the best sound quality I could afford. It took less than a minute to convert me to digital forever the first time listened to a CD!
Thanks for this video.
I remember how the first CDs had a high end scratchy sound. It was because the master was mixed that way because they knew the transfer to vinyl would attenuate the high end. So when they made the first CDs they just grabbed the original master tape and sampled it to digital. And the digital faithfully reproduced that extra high end instead of reducing it like the vinyl processing naturally did. End result was digital got a bad rep. Later they remastered with digital in mind and the next release of the same album was much better. So I ended up taking some of my early CDs and tossing them and replacing with newer versions.
@@kthwkr Ah, the (in)famous RIAA equalisation! That chapter gratefully closed forever, along with tracking weights, lateral compensation, surface noise etc that went with expensive turntables and magnetic cartridges.
The current infatuation with vinyl is beyond me.
@@kthwkrwas it like a crt whine?
@@jq4t49f3 Add to that the infatuation with tube amplification. I was there when the conversion from tubes to transistors happened and losing the distortion and bias of tubes was a big step forward.
@@ClareHehe Think of it more like turning up the treble control all the way.
As an audiophile I’ve always tried not to believe in things I don’t understand. I just experiment, listen to things and whatever my ears agree with is what goes. Many times I have preffered the sound of something cheaper, simpler or “less audiophile”. Or found no difference. The key is to pay minimal attention to what people are saying and just try things. Often they are just repeating things they hear or attempting to fit in with a group.
I am using lossy encoding whilst all the people scream FLAC. The amount of parroted nonsese is piling up to literal fairytales of untrue garbage. I play them my MP3 (!, the lowest denominator amongst AAC, OPUS, VORBIS and the likes) created with Lame 3.100 at V0 and they go like "Sounds actually pretty awesome". I tell them it's an MP3 averaging at 275 kBit/s and immediately they claim that they can hear "bad this" and "bad that", and that sound was "hissy" and how they rather want 800 kBit/s. I make an A/B switch on Audacity "showing" that there is no audible difference - nope nope nope, FLAC is the only way and MP3 is bad. Huh.... I even give them the WAV and all the "hissy" things are present in the WAV as well and not butchered by the encoder. Nah, dude, all bad. These people like to keep ultrasonic content and irrelevant audio information at all cost. . . . . .
Placebo effects are a thing, but not only with your ears, but also with your brain. Maybe analog to a kid that fell off of a bike: The knee hurts. Ouch, but fine. Same accident, same pain, but now the he kid sees blood on the knee. Waaaaaaaa!!!! Mama!!
In a shootout of high-end speaker cables a pair of wire cloth hangers won.
@@ZedekThat's the difference between audiophiles and audio engineers. Audiophiles rely on subjective impressions to judge their equipment, with all the human fallibility and bias that brings. Engineers have test equipment to measure it.
@@Zedek Yeah, back in around 2001 (I remember that year vividly because I did this at a unit I was living in when 9/11 happened), I had heard that LAME 256k CBR was supposedly so good that it was indistinguishable from the lossless source. I was running Debian Potato at the time and thought I'd try it out.
I started ripping with cdparanoia my favourite CD's and then converting those rips to LAME 256k CBR, to do some tests. I was doubtful, so as is human nature I was searching for unwanted, audible artefacts. I was shocked that these MP3's sounded so amazingly good and then suddenly I heard it! A warble in one of my favourite songs, which was not musical, seemed out of place and I did not recognise. I went back to the source CD and listened......... and there it was, that warble, on my pristine CD, of one of my favourite songs, that I'd never noticed before.
I checked other sources of the same song and what do you know, that warble was there too. I now cannot un-hear that warble when I listen to that song.
The ultra critical headspace I was in, while listening, caused me to find what sounded like non-musical artefacts in my favourite music, yet at the same time I was completely unable to find anything detrimental added by the LAME 256k CBR encoding process.
So yeah, LAME is a pretty awesome encoder and I understand it has likely improved even more over the past 20+ years and that there are some really awesome and ultra efficient open source codecs now (OPUS rings a bell).
@@ZedekDisparity of samplerate and audible lossy compression are not analogous at all, IMO. There are 100% people out there, including myself, who are not lying to you when they say they can hear mpeg compression. This is a vastly different conversation from the samplerate debate.
My best argument FOR the CD-standard is that it is not something they came up with and then proved to be good enough with math. The CD-standard was based on the proven math and was actually extremely stringent for its time. It's not more complicated than that with good electronics.
Things don't always work as well in the real world as they do on paper. Time travel has been proven mathematically but still not perfected.
@@acoustic61, well, that's true, buuuuut we HAVE empirically proven, as seen in this video, that the math holds for band-limited digital signals, and the same goes for the CD format.
@@TokeBoisen Music is more complex than some demo of a sine wave. I've listened to thousands of hi-res digital transfers and virtually every one sounds better than CD. I think it's easier to get better results with higher sample rates. Maybe because steep filters, which are imperfect, and other forms of processing can be used more sparingly. I see no reason not to use higher sample rares. Digtal storage is dirt cheap.
@@acoustic61 oversampling sounds better if any form of saturation is involved. So you're right that it won't matter for just a sine wave. Dan warrel has a good video on oversampling.
@@acoustic61, the sine wave is just the easiest example to demonstrate the concept. As is explained in the longer video from Monty, any complex waveform of a band-limited signal can be perfectly captured and replicated. If it deviates from the original waveform it MUST contain information above Nyquist and is therefore no longer a possible solution.
It HAS been documented and verified that phase-differences can impact the ability to capture perfectly if the chosen sampling-rate is exactly twice what you'd want to capture, but that is inconsequential for either CD or modern digital formats where the sampling-rate is much higher than twice the upper limit of human hearing.
Additionally, if you ever look at FFTs of hi-res transfers you'll most likely see that there is either no information above 20 kHz, or what is there is just noise. At best that means they just waste bandwidth, at worst it introduces IMD in the audible range. Any differences you are hearing are more likely to be due to a difference in mastering, an increase in gain, or just simply psychoacoustics.
Thanks for mentioning at the end that 24bit and higher sample rate is important for production.
Indeed, for recording one doesn't need more than 44.1 or 48Khz, however in production when there are tens of filters and effects applied to a sound, distortion and noise is amplified if the resolution is insufficient.
I can't think of anything that would create either noise nor distortion if the sample rate at which the audio is recorded at 44.1k. Oversampling for non-linear processing exists and for linear processing it's not necessary (and for things like convolution a higher sample frequency makes everything eat up computational resources. My DAW supports 16xOS, applicable to either plugin chains or single plugins, which is plenty. No need to waste CPU power and thus performance on using higher sample rates for things that don't benefit fromthem...
@@simongunkel7457 They produce ultrasonic harmonics, it you don't have space enough on your bitrate to accomodate them, they "bounce" back to the sonic range and they can be heard as "artifacts". So you need space where they can propagate, and can be cut clean when converting back to 16 bits, because all the "trash" is out there beyond the 16 bits resoultion.
They will explain this on the next video for sure.
@@framegrace1You are confusing sampling rate with quantization (bit depth). The bounce back happens if you have non linear processing (like all distortion) and the sampling rate is too low. Quantization only affects the noise floor.
@@framegrace1 You are confusing bit depth and sampling rate here. There are good reasons to record at 24 bit depth, but that doesn't have to do with ultrasonic harmonics, but it allows for more headroom when recording and more tracks to be mixed (you lose about a bit for each quadrupling of tracks, so if your project has 64 tracks that will mean your output will have lost 3 bits compared to the recording. And you also lose a bit for each 6dB of headroom while tracking. When 16 bit was state of the art, you had to track really hot, which made clipping likely. With 24 bit, you don't have to track hot at all and won't clip and still have plenty of bit depth to allow you to go ham with multitracking). Aliasing, i.e. ultrasonic harmonics bouncing back from Nyquist has to do with the sampling rate. But as I mentioned you can use oversampling to deal with that. So my recorded audio at 48k goes to a non-linear plugin. It will alias. I enable 16x oversampling for that plugin, which will make the plugin see a 768k signal where the ultrasonics get to live up to 384k and then have another 360k before they would bounce below the 24k where they could end up in my 48k signal. That is then filtered out and the signal is converted back to 48k. But if I send my 48k signal to a linear plugin, it won't produce any harmonics and thus I don't need oversampling and will just run the 48k signal. Convolution is usually the most intense processing and it is linear, but the computational load scales with the sample rate. Note that my line of thinking here requires a DAW or plugins that oversample. I'm using reaper and per plugin oversampling has only been a feature for less than a year and not every non-linear plugin has internal oversampling (though plenty do). But even then tracking at higher sample rates rather than processing at higher sample rates didn't make much sense and now you can get granular with that. If you hit everything with 96k, you will slow down things that don't produce aliasing anyway and things that do will still have bigger issues than if they specifically got 768k signals to work with.
Exactly, processing in the digital domain will degrade the signal, unless you start out with a higher quality already. Just a simple +3dB filter already throws out one bit of dynamic range, so you better start out with some headroom.
Somehow audiophiles got wind of the fact that these audio formats exist and that it would be better to listen to this original mastering instead of a downsampled copy, which obviously makes zero difference, unless you can upgrade your ears to higher dynamic range and higher frequencies.
Although I'd still say that for typical audio mastering, only the bit depth will actually matter, unless you have some serious slowing down planned. Otherwise the high frequency components would remain outside the audible range either way. Maybe it might protect better against some quantization artifacts when filtering.
I worked for BBCTV for 27 years as a sound operator and was there when digital recording reared its head. This video is most interesting and actually rather reassuring. Thank you!
How can we get the English version of "The Flying Dutchmen"
One of my favorite BBC R&D inventions is sounds-in-sync that used the same concept later used for digital recording with PCM converters to videotape, by digitizing analog sound and then converting that binary signal into a black-and-white television signal that could be sent along the standard black-and-white signal. The BBC used this also to feed FM transmitters where the digital sound would only be converted back in to analog and modulated onto FM in the transmitter. Basically Britain had the pre-cursor to DAB radio running from 1970 onwards - pretty impressive.
*Insert MJ in Thriller eating popcorn gif*
Too bad GIFs aren't allowed on TH-cam! Thanks for watching, John.
@@AudioUniversity many things are not allowed here. Better go to concurrency.
@@AudioUniversity TH-cam does not allow a video of someone doing a finger stick test to check their blood sugar, the drop of blood might offend or scare someone...sigh.......
As a member of the general public without any background in sound engineering who always heard of the stair-step wave thing and how digital sound is "worse" than analog because of it... Your explanation is SUPER on point. I was able to fully understand the concept. I give you kudos for being able to create content valueble for specialists and laymen alike. You're awesome! :)
Me too! But I had trouble understanding some points. For instance, I've never heard of the stair-step at all, so I had no idea what it even was, honestly.
Yeah, but he's wrong.
@@toomanyhobbies2011 elaborate
@@toomanyhobbies2011 well, he presented a long video with a lot of sound, reasonable argumenting for his point. You'll need more than "he's wrong" to be taken seriously. So please elaborate.
The people that tout "worse than analog" usually compare the most expensive analog solutions to the cheapest digital ones, in which case you're just comparing the relative quality of the equipment. But similar goes for the other side of the argument, if you compare a high-end digital solution to a bargain bin analog one the more expensive one will objectively be better, but again not because it's digital.
If you get something decent and keep the signal properly intact from start to finish both will be as good as the other, with some minor changes in the tuning depending on what all is in the chain at what point. I have both and enjoy using both depending on what I want to listen to at any given point.
Analog is a bit more prone to outside interference like ground loops or cables picking up noises from a nearby power line, which attracts purists that want to feel better about themselves using a harder medium, but digital is much more convenient and thus easier accessible to everyone.
Good music should be able to be enjoyed by everyone, not just elitists.
Thank you for uploading this! I saw Monty's video over 10 years ago, but just a few years ago I tried to find it again and couldn't. This video came up in my YT feed and something told me it might be what I was looking for.
Higher bit depth though is used in the production process to be able to do things like adding two loud signals or other processing functions that would otherwise add quantization noise (ever see the bands in a dark part of a streamed movie?) or blow past the highest signal of a lower bit depth format. The signal is then requantized before mastering back down to 16 bits. The creators of the CD format really knew what they were doing, and had over a century of digital and signal processing research to call upon to come up with the 16 bit/44.1 kHz format.
Same thing for sample rate: They have a use in production. When you want to do non-linear filtering without the higher harmonics getting aliased back down, or because it helps to avoid filtering artefacts, or so you can slow down a sample without it turning really base-y. But those are only intermediate stages, needed to help the mathematics of transformation work. Once production is done, it's all turned back into something more suited to limited human hearing. 44.1KHz, or sometimes 48KHz.
@@vylbird8014 Exactly!
All true. High sample rates and bit depths are very important for production. But it's a complete waste for distribution. But it's not ALWAYS important for production either. Aliasing is not as common a problem as most people think. Not everyone is pitching and slowing recorded content down. And as for the higher dynamic range - it's very helpful to have that headroom in a recording, but again, is often not an issue. A lot of producers use it as a safety net, even though they rarely need it.
I'm more excited about the move to record in 32-bit float. That's not about dynamic range, but about never clipping unless the mic itself is overloaded, and never having to worry about gain until you are mixing. It's like the audio equivalent of shooting photos in raw.
@@mwdiers That would be really cool. Eliminate an electronic stage for gain adjustment in recording?
I thought the 44.1khz, 16 bit was due to Umatic tape restrictions?
Fabulous. I have understood this although at less technical level until now. That testing method he shows is absolutely brilliant and his explanation is spot-on. Those first engineers at Phillips and Sony who created this standard and the Red Book, they knew what they were doing and set the bar at a very high level. This is a fantastic addition not only to your content library but for the general public.
Thanks, @Grand Rapids57! I agree - the bar has been set incredibly high. Unlike developments in video recreation, audio has really stood the test of time since the CD.
No wonder - although the result was wonderful ! Audiophiles were much more inclined to testing equipments… And, what’s more, the human ear is an outstanding tool, much better at discriminating than the eye : visual illusions are plenty, but aural illusions are just a few. And if you want to know the material of a wall, your eyes will easily mess up. Knock it with you fist and you’re ears will tell ! 🤓 Hearing is knowing, as exemplified by the fact that no one will make fun of a blind man, but a semi-deaf man is a funny character because he « doesn’t understand » and keeps making stupid mistakes… In French, « entendement » is what translates as « understanding », suggesting that « entendre » (I.e. to hear) is intimately linked to the knowledge of the outer world. Outstanding video, by the way. Deserves much more views ( sound included), even with the sloppy YT compression !
Philips wanted to stick to 14 bits in the beginning and only a short period before the market introduction the specification was changed to 16 bits under pressure of Sony. The first Philips CD players used 14 bit D/A converters and they implemented oversampling to end up with 16 bit accuracy.
Philips already had the design of their DAC’s ready for production so there was no time to create a new design so the clever Philips engineers came up with a 4x oversampling design using the 14 bit DAC to end up with 16 bit resolution.
Harder than Phillips stick with that before switching their stuff natively to 16-bit? Did it have a bad effect on their sound?
@@actionjksn they couldn't stick with 14 bits since the CD format was already determined to be 16 bit at that time. Using a 14 bit converter without oversampling would have caused degredation.
In the end it didn't cause any downsides in terms of sound quality. Philips was also on top of there game in that period, they produced some of the best DAC's in that period.
Thank you for this amazing video. As an old film sound engineer I used audiotape even without Dolby noise reduction, optical sound, mono stereo, and later digital systems. The most remarkable experience was a CD, a classic concert with orchestra, choir and organ, replayed from a commercial Philips CD player and after that replayed from a similar CD player where all electrolytic capacitors were replaced by other capacitors and copper leads and wires were replaced by carbon leads. The recording sounded much better and you had the idea that you were in the recording place yourself. The engineer who constructed this CD player was clashing with his bosses and with Philips to keep this invention quiet.
If you want to hear a good example of quantization noise, the Game Boy Advance has digital sound channels that are limited to 8 bits. Accordingly, many games sound like they have tape-like hiss, but only when sound is actually playing.
Key to this is limiting the bandwidth of the sampled signal. With proper low pass filtering, frequencies above Nyquist are eliminated and there will be no aliasing noise. However, oversampling can simplify the input filter complexity and the high frequency content (noise) can then be filtered digitally and downsampled to generate the output digital stream at 44.1K or 48K samples per second.
Internally audio production software uses 32 bit floating point to avoid clipping the 65536 values a 16 bit software synthesizer can generate. The final output is 16 bit though. 32 bit floating point is also faster for cpu calculations. In this video only the end results are matter-of-fact for human hearing.
There should be aliasing at 15 khz. Even if you generate the perfect waveform on the PC, there are clearly audible aliasing byproducts.
@@dtibor5903 2 times 15khz is 30 khz, but TH-cam cutoffs everything above 16 khz fo better recompression.
That's why it's 44.1K rather than 40K - need a bit of extra room because real-world filters are not perfect, you can't have a brick wall. Well, you can with oversampling and digital filters, but that wasn't around back when CD was introduced.
It's actually the output filter that restores the original analog waveform. So the quality and steepness (or rather the lack of steepness, and here come oversampling and dithering into play) of the filter are essential. Maybe it should've been mentioned in the video.
Edit: Oh, I see Max Nielsen has already explained this in a much more professional manner. So it is with those who write before they read.
Agreed. A lot of people say that it's not a stair step, but the digital chain goes Dots from the PCM file -> Stair Step at the DAC chip -> Smoothed out using a filter. Source: Texas Instruments, maker of DACs
Many people have not grasped the entire chain and seize upon denouncing the stair step saying it does not exist. It might not exist at the speakers, but deep inside, it does exist. That some people fear the stair step at the speakers is perhaps reasonable in the case of an extremely cheap system that makes no attempt to filter and uses 8 bit, 11 KHz sampling which is barely adequate for voice and the sampling artifacts are clearly audible unless heavily filtered, in which case the heavy filtering is noticeable.
@@thomasmaughan4798 Audio University has given out so much misleading information that has led to online arguments that someone needs to do a debunking video. This video title is very misleading
@@thomasmaughan4798 And where do you buy 8 bit, 11khz discs?
Agreed.
Some of the reasons to sample higher than 44.1 kHz was to make anti aliasing filter less steep to avoiding aliasing. This filter is in the frontend and analogue, so the filter is cheaper and gives less phase distortion. If the signal is oversampled many times you can take advantages to noise shape the spectrum (place the noise in the frequency range you will filter away).
Sigma delta converters do just that.
@SR-ml4dn In the original article, Monty does mention that *processing* at higher sampling and depth is perfectly sensible, the original article (which is linked in the description) was a response to "high definition audio". And the video here is a complement to that, clarifying misconceptions about digital audio & audio waveforms.
Front-end analogue filters are a problem because they can't be sharp enough without introducing distortion and noise. Instead we just sample at, say, quadruple the rate (176kHz) so the analogue aliasing filter can be much less sharp. Then we use a digital filter before down-sampling to 44kHz. Even digital can have trouble with the sharp cutoff between 20kHz of hearing and the 22kHz nyquist limit. Moving the nyquist limit up to 24kHz helps which is why 48kHz sampling is so common in newer standards, including digital video formats.
@@rogerphelps9939most/all silicon based sigma delta converters create idle tones and some cause noise floor modulation. This is why engineers like Bruno Putszeys designed discrete PWM conversion methods.
Came here to say this having a masters degree in vibrations for mechanical systems. Aliasing and side lobe distortion is always a major issue when recreating frequencies from sample data with FFTs and I see no reason why it would different here.
I remember the origins of the CD and people dismissing digital as chopping the sound into little bits, that just has to sound awful! But of course most audiophiles are not that technically educated. Maybe just the ones that design and build the stuff. And their opinions are thus also tainted by wanting to sell you something.
"But of course most audiophiles are not that technically educated."
Whereas you are wise and smart.
When someone spends a lot of money on something, they'll do anything to convince themselves it was worth it. It's all that is!
@@thomasmaughan4798 Does one have to be all that wise or smart in order to be technically educated in a topic?
(In case there was any ambiguity: I don’t mean this as any kind of dig at people who are technically educated in the topic)
@@drdca8263 Some people are offended when their ideas are proven not to be possible.
I didn't really think CDs would sound good when I first heard about how they work, but I also didn't think of myself as some kind of superexpert on all topics so I just listened to a couple. I had friends who swore the quality wasn't equal to vinyl but I couldn't hear any difference other than the vinyl was noisier. So... I bought myself a fair number of CDs.
I learned this while studying computer science. It was one of the basics required for the bachelor. The same methods/theories are used for all kind of signal processing like WLAN or cell phone networks.
Thanks but I prefer analog WiFi. Reddit just has a better more real feel to it.
it's also used in television, radio, basically, if it has a signal, it likely at some point in the signal chain, uses the logic of the Nyquist sampling theorem in it's design.
is that why my cell phone sound so good?
@@oldolfmann8927 It's why the most common word in cellphone conversations is, "What?"
@@davidryder3374 my comment was sarcasm LOL. Digital does have it's place, but there are a lot of places I do not like it.
I knew this had to be about Monty as soon as I read the title. That video was seriously one of the best explanations of digital audio on the internet.
I would guess they use 24 bit high sample rate in music production for essentially the same reason movie producers and Photoshop artists use significantly higher resolution source material than the final product, because when the final rendering is done, any subtle errors in production will be ultimately diminished to the point they are barely if at all noticeable
Exactly, it's about preserving dynamic range of the finished product. It means you don't need to perfectly hit full scale range when you digitize the audio signal. You want to leave a bit of headroom when you feed any analog signal into an ADC so clipping does not occur. If you record at 16 bits and your resulting signal only uses 15 bits because you left yourself just a little headroom to avoid clipping, you can never get that bit back and your product can not be any better than 15 bits deep, or about 90 dB.
If you record at 24 bits and your signal only comes in 20 bits deep because you left some headroom that means you have 120 dB of dynamic range so you can hit your 16 bit/96 dB target.
Also, if you process everything at 16 bits then any added noise in the processing chain means your finished product cannot meet the 16 bit/96 dB dynamic range target of a CD.
I'm not sure exactly what advantage a sampling rate about 44.1 kHz would give you.
Yeah. I have a little pocket field recorder and I record at 24bit, mostly because then you don't have to care much about recording level - as long as it's there and it's not clipping you just adjust the volume in post
@@Fix_It_Again_Tony I'm not an engineer but a guess about one reason of probably multiple reasons why they might sample at 192k: When a studio records a track, each instrument/voice is recorded separately. Each track has overtones and such that have frequencies above the 22k limit. But when you mix them all back together, you end up with interference/resonance that results in some sound being audible to human hearing. A small, almost imperceptible part of the audio that you would hear if the band was all playing live in the same studio room
@@Sumanitu no, 192k is because nyquist thing, with analog filter you will never remove frequencies > 24khz and that will damage signal bellow 24khz during sampling. So low order filtering, 192k sample, then hi order (almost ideal) digital filtering to get rid of everything above 24khz and then resampling down to 48k
@@tkupilik the "nyquist thing" is literally what I'm talking about, bud. You dont want to lose frequencies above 22khz (1/2 the 44khz sample rate), even though they can't be heard by human ears. Thanks for re-explaining exactly what I was talking about...
I have lots of different music formats: SACD, MP3, FLAC, DVD-Audio, Blu-Ray, and on and on. This video is in line with my observations. The original recording, regardless of the format, is much more important. But I still like high-density formats. I cannot tell the difference between ANY of these formats, EXCEPT when I crank up the volume. I can crank up the volume on my giant speakers maybe 5-10% louder (which I love) without distortion with high density formats (and I have plenty of the same recordings in multiple formats). But I always assumed it had more to do with the way the sound is processed by my specific electronics than by the actual bit rate. It varies by specific systems. This video implies that the noise reduction of higher bits may be the culprit, but given all the electronics issues involved, it's hard to pinpoint. Same with vinyl - I can get some unbelievably sweet sound from certain vinyl that just isn't the same from digital, but only with excellent (but rare) specific records, and only with crazy complicated setup and very specific components I've discovered via long-term trial and error. So again, it seems to have more to do with the specific setup than the actual music format. I love vinyl, but I laugh at the complexity of getting truly great sound with it - so many variables.
Thanks! It seems that we are reaching a point where younger generations are getting educated on these topics, and ruthlessly debunking all the old myths. I'm enjoying every moment of it. The same thing is happening in the world of electric guitar. Shout-outs to Jim Lill.
His videos are awesome, seriously. The guitar tone one blew my mind
It's hilarious to see guitarist's creating ridiculous theories to "debunk" Jim Lille's vids to justify buying their guitars for the tone wood
It's not an "old myth", this video is in error.
Exactly! Makes my day to see Jim Lill mentioned here too.
@@wizrom3046 how so, please explain.
I remember watching that way back in the day. That original video was excellent, and it's a message that definitely bears repeating.
I've been looking that original video for ages since I saw it years ago. It clears up so many myths, especially he confusion between sample points (infinitely small) and the "fatbits" version audio software tends to show (though nowadays some actually show the equivalent of what the DAC will output, with proper curves).
That bit about pixels around 6:56 clarifies some confusions I always had about pixels when trying to contrast them with vector graphics
Arguably, that's bullshit too. Whatever you used to capture the original image DIDN'T have any sort of idealized pinhole-sized pixels - it had a surface area of some shape (actually rectangular most likely), and what it captured was the sum of all photons landing ANYWHERE WITHIN THAT AREA. Representing that as a square IS the sane way of saying "withing this specific area, this is the color we captured, and we have no finer-grained information concerning that area". And that remains true either for digital cameras or rolls of film - only there your "pixels" are the crystal grain of the film, of irregular shape...
Licensed professional engineer with undergrad degrees in engineering, physics, and mathematics and a masters degree in engineering AND a professional music producer here. You just got a subscribe. This was one of the best videos I've seen in awhile... and though I theoretically understood this based on DA design concepts, this video spelled it out with such clarity and reason, that I had to comment.
Great work!!
Just get rid of the lollipops. The output of DAC is latched, there are no lollipops and the staircase is real -- just microscopically tiny and passed through a filter to get rid of the 44.1 KHz sampling frequency.
@@thomasmaughan4798 When it's reconstructed to analog, it definitely isn't a stair step. Look at the output on a scope. Show me the stairs steps.
@@mc2engineeringprof " When it's reconstructed to analog, it definitely isn't a stair step."
It most certainly is! DAC can ONLY output DISCRETE voltages. You give it a 16 bit code, it spits out the corresponding voltage until it gets a new 16 bit word. Then the voltage jumps immediately (within a few nanoseconds) to the new voltage.
"Look at the output on a scope. Show me the stairs steps."
You CANNOT see 65536 steps on a scope with barely 8 bit vertical resolution! Noise will exceed the stair steps anyway.
Now, if you increase the vertical gain to zoom in on a *portion* of the sine wave, and you probe the output of the DAC itself, you will see staircasing. *That is how it works*
Some TH-cam videos compare oscilloscopes on this exact procedure. A digital scope creates staircases on *input* depending on its ADC. That's why you use an analog scope for this sort of thing; a digital scope itself introduces staircasing.
But what is the resolution of an analog scope? It is noise limited for one thing; 16 bits is a LOT of depth. A typical CRT for television has 512 lines of vertical resolution, it is impossible to see 65536 stair steps when the phosphor dot is already straddling hundreds of these steps.
@@thomasmaughan4798 You don't understand how D/A conversion works. Watch the video again.
@@thomasmaughan4798 you're describing audio DACs from the early 90s maybe, they haven't worked like that the past 25 years though. Modern audio DACs use upsampling and sigma-delta modulation to convert the PCM to very high sample rate (a few MHz) but only a few bits of resolution, so the unfiltered output may not look like the original signal but it also does not look like the kind of stairstep you're describing. More importantly, it's kind of irrelevant what the unfiltered output looks like, the external analog filter is essential to the proper operation of the DAC. The overall result is that it will properly reconstruct, to a high degree of accuracy, the (unique) bandlimited signal that passes through the sample points given to it, provided that this signal is within the passband of the DAC (which is at least 20 kHz when using a 44.1 kHz sample rate). The details of how exactly this is achieved is quite interesting but not particularly important to understanding digital audio as a concept.
i love how a more or less old video explains is SO well and still people don't believe it and keep repeating confusing or flat out wrong info
Audiophile is just another word for superstitious, really. Never met one who had any technical knowledge at all.
*proceeds to change all settings on my audio drivers to 44100 CD and leaving them like that forever* It helps so much with eliminating higher processing needs that might end up in artifacts such as stuttering or crackling of the samples. Thank you! This has literally changed how I use studio equipment.
There is a benefit for using higher sampling rates, but it only applies when making the actual recording (I think this is what is alluded to at the end of the video - looking forward to the next video to see if I'm right!). Although humans can only hear up to 20 kHz, physical sound waves include frequencies higher than that. Those frequencies must be filtered out before sampling or else they will get aliased into the hearable range. But no filter is perfect and frequencies above 20 kHz will pass through the filter, though higher frequencies will be attenuated more than lower. So the solution is to sample at rates much higher than the filter's baseband bandwidth so that only the highly attenuated very high frequencies alias into the hearable range.
That's exactly the point. Ans also why amateur recording and reproduction of high bitrate without modification can actually sound worse than lower bitrates. Those are the ultrasonic harmonics that he talked about.
Well if you record a 20khz sound at 20khz, then every sample will get the wave at the same position, so the computer will assume it's just a straight line. So you need at least double that to make sure you can get both the peaks and troughs of the sound wave.
@@goldenfloof5469 Yes, however Nyquist is a double-edged sword. A 30kHz wave recorded at 20kHz will look like a 10kHz wave; that's why sampling high and then digitally filtering can produce nicer audio.
This is called "oversampling" and it's built into every modern analog-to-digital converter, even the cheap ones. There's no point in recording at a high sample rate and then downsampling; the converter does it for you.
@@allochthon I am an electronics design engineer; this is simply not true, sorry to say. That being said, you may be right with respect to half-decent converters which allow for multiple sample rates to be set. The problem is up-sampling and down-sampling with high fidelity uses non-trivial algorithms to implement in hardware. 48kHz -> 44.1kHz is an ugly conversion, for example, and a surprisingly common maximum for inexpensive electronics (though 96kHz is becoming more common). You can't just average data when down-sampling and get the same results as oversampling and then software processing.
I record at 44,1 to avoid downsampling, but use 24 bits to provide more headroom for effects plugins in each track. After mixing I export to 44,1/16.
Yes. There are several reasons to use higher bit depth and sample rate for audio production (recording, mixing, and mastering). That’s what the next video is all about. This video is just about audio playback.
@@AudioUniversity yes, for playback 44.1/16 is perfect. My CD collection is still growing and I really enjoy how good they sound considering how affordable a great sounding CD player is nowadays.
24bits does not give you more headroom. Headroom is loudness or volume above Odb(VU) or your normal operating level volume!
24bits give better sample accuracy or sound dynamics so does mean you are sending a better (more accurate signal to your effects)
@@AudioUniversity waiting for next video
@@rods6405 Depends on how you transform. You can easily have more headroom with 24bit and use it for expanded dynamic range, this is in fact one of the the primary uses of it before your compress it to 16bit at selected boosts.
0:15... actually, this goes against what the "I paid $2000 for my interconnect" audiofools say. The actual experts, the people who actually know about this stuff, sound engineers, acousticians and the like, completely agree with you.
A HUGE thank you! Excellent presentation! The Philips/Sony engineers that developed the CD format were brilliant!
I was in college getting my engineering degree while Philips/Sony was developing the CD format so most of this information is NOT new to me, but what was truly mind blowing back then was storage density. My buddy and I were going through the College of Engineering junk pile and found two MASSIVE hard drives, and having just heard what the CD format was going to be, we started crunching numbers to use those junk hard drives and other instrumentation to kludge up an A/D and make our own CD format files from taped live recordings. Well, the whole effort ran out of steam when we figured out that the MASSIVE hard drives could hold at most eight seconds of music. Eight seconds.
Oh, and I'm still using the Philips CD player with TDA1541DACs that I bought in 1988.
Good discussion! I stand by my assertion that the best improvement in my stereo was the cheap CD player I bought back in the early 80s. Gone were the days of clicks and pops of vinal! It is interesting that vinal is making a comeback. I think it's the ritual of the turntable and cleaning that's the reason. I did double blind testing on very high-end equipment from the late 70s and early 80s and the CD always sounded better. The only thing close was a $5000 turntable/cartridge combo that was awesome! Sounded almost as good as a $130 CD player with a DDD disk.
My best takeaway is it's the music that's important and not the gear! You're far better off spending your money of good quality recordings than expensive equipment.
That said... Spend money on good speakers. Only the microphone (which you generally don't control) and your speakers should color your sound. Tube amps color sound in a nice way (even harmonics) but at the end of the day, it's still distortion! Great for guitar amps! And yes, I'm listening to this on an amp with a tube preamp because "tubes"! They are cool and glow! And have a VU meter! And that's through speakers of my own design. I'll switch back to a cleaner amp soon. But the VU meter is cool!!!
Thanks for the video!
Higher sample rates are great if you are a sound designer and wish to do lots of time compression and expansion, particularly if you wish to get fewer digital artifacts. There are also times when you want the digital artifacts from time compression and expansion, so it’s all about that your desired effects are. Knowing when and how to use each audio sample rate and bit depth is a skill you get from experience. Great breakdown on the technical side for those not in audio production. Can’t wait to hear your thoughts on how to use each sample rate and bit depth when it comes to professional audio production. I wonder if you will get into more than just music production, but also pre and post production for TV/Film/Radio/Podcasts, Sound Design, Foley, Voice overs, and Voice Acting to name a few…
Mixing down to 44.1khz 16bit is perfectly fine and sometimes preferable, it’s only when you start messing with audio’s time and pitch do you really start to see immense utility in higher sample rates like 196. It’s great as a Sound Designer when you get to choose between different sample rates and bit depths, it adds so much flexibility when tweaking sound effects and voices. It can save lots of time and money if you have the ability to tweak sound effects and ADR VO to work and not have to record new sound effects, or especially bringing back in an expensive Voice Actor just because you are limited because of sample rate just how much you can manipulate a particular voice/sound effect…
Just my two cents…
Off course high sample rates matter when it comes to sound manipulation in samplers/DAWS/vst-i:s. But his video was only about sample playback for a fixed song with a wave that isn't manipulated. Even back in the 85 when the Fairlight CMI 3 was released, they knew that playing back 100 KhZ samples pitched down an octave would get you really cool sounding low tones that still retained "full quality" with less artifacts!
Oversampling is also used in many softsynths when they are doing "internal calculations".
So much disinfo here I don’t know where to start. Without realizing it, you are propagating the very silliness this entire video was made to debunk. I suggest you study dither until you actually understand it for starters. Then do some more reading to find out how top engineers actually do their digital recording. Sound Design is a somewhat low demand version of music, not the opposite. I’m happy for you to feel special about your chosen line of work, but it would behoove you to know what you are saying before you write these ridiculous screeds.
It is a pity that the embedded video did not explain why the step function is not the end-result of digital analog conversion: namely that the signal is subject to subsequent thorough low-pass filtering. Without this information the nice harmonic output of a DAC is surely quite mysterious.
Yes, thank you
Yes, the missing reconstruction filter has been mentioned in the comments a few time. That it is there is indisputable. That it is not mentioned in the video in my opinion is most likely due to the video being a digital audio marketing video from back in the day.
Even the essential anti aliasing filter on the input is not mentioned probably for the same reason, to prevent the viewer from being horrified at having "filters" in the signal path.
Other than these simplifications it is a remarkably good demonstration and reflects the real life experience of digitised audio.
@@KallePihlajasaari As for me, I am delighted to have filters in my signal paths. They are beautiful results of human mathematical thinking.
Nyquist shows that the stairstep must be converted to Dirac delta conversion to simulate discrete time before lowpass filtering, so it's not the filter, it's the discrete-time approximation.
@@tomgroover1839 Huh? Anything can be filtered. The stair step is the default output of a DAC. Other digital filters may be used and these days often are but a low pass filter will remove higher frequency harmonic noise due to quantisation irrespective of what it looks like in the time or frequency domain. The point is that the lollipops are a representation for the mathematics on paper and are not used electrically. The electrical representation is the stair step until filtering takes place which typically is used.
Apparently some people won't accept the Nyquist result no matter how much evidence you show them.. It seems to be almost a religious conviction. The same people probably have very little ides of what a Fourier transform is. I guess the only way to (maybe) convince them is to put them through a double-blind test situation in which they have to prove that their 'magic ears' can discern the difference. This has been done numerous times, I'm sure but I don't have any cites to hand.... do you know of any good links?
“Almost” religious? I would say it’s decidedly so! :p
Glad you mentioned the issue with trying to play output that contains ultrasonic content may cause distortion. Many speakers will not have the frequency response to reproduce signals with ultrasound components as they were not designed to and they are very far from the frequencies they were optimised to reproduce best. This will usually be 20Hz-20kHz for a single driver speaker. The reason why no speaker can reproduce all frequencies equally well is that the strength of the electromagnet has to be tuned with the mass of the cone, the range of travel of the cone and the desired frequency response in mind. Play frequencies too low through it and the cone may move too far and physically clip out, play frequencies too high through it and the cone can't really respond quick enough for the smaller fluctuations in the signal producing noise in the mechanism that way. This is also why high end speakers often have multiple drivers fed through band pass filters, the big heavy driver that can push lots of air for that thumping bass is really not suited to producing tones in the higher range that need a more nimble cone or more powerful driver. The problem is putting in a driver that can move that monster through it's full range 15000 times a second and it will hit the limits of it's range of motion part way through the cycle with a 20Hz wave of the same amplitude. On the other end optimising it for a few hundred Hz means the cone doesn't have time to move far before the signal switches at high frequencies. So you can't have your cake and eat it and trying to demand a speaker to do everything will make your listening experience lousy.
I've been evangelizing Monty's video (and its accompanying article) since it came out. Glad to see someone else spreading the good word!
Fine work sir! My argument for all the vinyl lovers has always been db range. The maximum for LP was around 60db, not including noise floor. Then add in the pops and crackles after undergoing the RIAA standard. CD has been a great playback and preservation method (so far) compared to vinyl or tape.
playback yes, preservation not so much.. i used to work for an archives compagny and we prioritize CD over Vinyl and tapes, as the support is way durable if conserve in proper conditions. Average life time for a CD to start degrading is around 10 years if i remember correctly. It is way more for others medium
@@mambocountach I've been collecting CD's since the 80's and none of them have failed whether stored in a cool place or warm storage facility. Tape and vinyl don't do well in heat. I say this after retiring from the record industry after 30 plus years.
@@mambocountach I recall there were some CDs manufactured in the UK in the '80s that began to delaminate after 10 or so years.
Here in Australia a lot of early CDs were pressed by Disctronics in Melbourne, and to my knowledge none of those discs in my collection have degraded.
@@mambocountach My thinking is that there is no lifespan limit if the disc doesn't delaminate or get fungus(?) between the layers, a very rare phenomena. I have 40 year old Hi8 tapes that play like new, despite the experts claiming a much, much shorter lifespan. Supposedly, stored tightly wound the magnetic fields from one layer will imprint through the substrate to the next, mixing things up. Yeah, right.
And vinyl has a maximum bit depth of 12 to 13.
I'm very happy to see people start to explain this to the masses. For 2 decades I've been telling audiophiles (extremists?) that they rarely have less than 2-3 bits of noise on the analog part of a 16-bit sample and they don't believe me. Even explaining that 16 bits means 16 microvolts of resolution per volt and that it's easy to get more due to RF radiation around. Not to mention that some such people nowadays use class-D amplifiers and find them good while these ones generally provide less than 8 bits of resolution due to using an insufficient sampling frequency! Thanks for this video, really!
I love the Digital Show & Tell with Monty! Every once in a while I re-watch them (parts 1 and 2) to freshen up my knowledge around the subject :)
Great video! I want to clarify something about the Nyquist rate: it’s not necessarily that you need to sample at twice the highest frequency, but twice the bandwidth. If your signal of interest is, for instance, between 800 and 1200 Hz, then you can apply a bandpass filter that accepts that range and rejects what’s outside of it. With such a band-limited signal, you don’t need to sample at 2400 Hz, but 800 Hz! Essentially this technique exploits aliasing to reproduce high-frequency signals.
Source: I work with RF professionally
Great post! I wondered when someone would correct the faulty definition of the Nyquist theorem. I have also sub-sampled many signals in my days. Also, in the video it was stated that the stair step view is never there. You do get a stair step when you have no DAC anti-aliasing filter.
This is not entirely correct. By this logic a signal between 1GHz and 1GHz+400Hz would only need to be sampled at 800Hz. You're missing the downcoversion step -- mixing with the carrier and low-passing.
@@dmitryjoy Actually you can, without down sampling, as long as the input stage in the ADC has a sufficiently large full power bandwidth. Now you have to bandpass it to only let through your band of interest first though. Sampling at 800Hz with a 400Hz band of interest would be an impossible filter to make since it has to be a "brick" filter though. You also have to be smart about what sampling frequency that you pick. I have done sub-sampling implementations in a professional setting, with RF signals. The linked video explains it. th-cam.com/video/ryJPVHrj0rE/w-d-xo.html
So true, however with CD Audio the dead bands below 20Hz is not worth mentioning as a credible saving.
In early A-Law and u-Law phone service codecs they used 8ksample with input filters of 200-2800Hz giving just 2600Hz bandwidth that in theory could have been achieved with 5.2ksamples with perfect fairy dust input filters but simply not worth the trouble for the savings.
The input circuitry on VERY high bandwidth oscilloscopes makes use of similar techniques elevating them to near magical levels with parallel sampling of delayed signals as required to compensate for the inadequate sampling rate but still maintaining the input bandwidth. So a 2 GHz input might use 4 x 1GSample ADCs or 8 x 500MSample ADCs plus a lot of DSP magic.
This was a great video. I am a retired electrical engineer. Most of my career was concerned with DSP (digital signal processing) of one form or another. I think that the reason that digital audio mystifies people is that you must have an understanding of some fairly advanced mathematics to understand why it works. Topics such as Nyquist sampling theory, anti-aliasing filters, spectral images, and Fourier transforms come to mind. Most people attempt to apply their common-sense analog thinking to it but that just doesn't work.
What a great video! I was aware of the technical inaccuracy of the stair-step, but I'd never seen someone show the analog signal converted back through an oscilloscope to show it that way. Very interesting.
Thank you so much for making this educational material available under Creative Commons, by the way.
I just felt a great disturbance in the vinyl market. As if millions of hipsters cried out that record players are better and were suddenly silenced
I buy vinys as fancy containers for mp3 downloads. They are nice display pieces.
I still like the sound of a good well preserved record. its the coloration which seems to have that ear tickling effect, not because its more accurate.
Vinyls still sound great, but it's more because of how they are mastered different from CD than anything to do with it being analogue. Can't have a vinal with so much range that it jumps your needle off the groove and all that.
@@lorestraat8920 Part of what makes valves so appealing to the ears is the harmonic distortion they produce. Its a pleasant effect. And vinyl has a similar effect provided the mastering is done well.
But where i truly thought where the debate would hold water, was when recording live instraments to my Teac Simulsync with ampex 456 on 15IPS. Vs recording them to a really good ADC into a computer. The ADC i have is Universal Audio equipped with all Burr Brown opamps and other quality parts. About as good as ADC can get. Using the same mic and mic pre, the Teac sounded better to my ears every time. Its like it preserved a certain warmth and depth that the UA ADC couldn't, and I even tried other recording interfaces as well.
Now having the UA ADC in a 64 bit floating point and high sampling did manage to match the Teac in that respect... But Otherwise I am unsure why the analog equipment was seeming to preserve the sound in a more pleasing way.
Its why I have tended to keep my recording in the analog domain while it makes sense to do so. Provided it is preserved without degradation from generational copying.
If CD really was as good as these people say then it would have killed off vinyl long ago. The problem for CD is that most musical instruments emit ultrasonic frequencies above 20khz. The CD just cannot reproduce those frequencies, but vinyl can. Also, these frequencies have been shown to have a positive effect on humans. But they don't want to talk about that because they are not particularly smart engineers. But ask the engineers who work for Sony and you will get a different perspective because they actually use their brains.
I became interested in electronics when I was a kid during the mid 1970s. I discovered high fidelity audio as a pre-teen during the late 1970s. At one time, my dream job was to become an EE and work for McIntosh. Then during the early '80s, I added computers to my list of passions. In college, I earned a degree in computer engineering, but I also took all of the electrical engineering coursework as well, which covered things like analog circuit design and analog/digital signal analysis/processing. My professional career has been in computer networking, operating systems, and embedded software development. But I never lost my love for (vintage) analog hifi. (It just took a bit of a hiatus when I was busy raising a family. Now it's back in full swing.)
From my educational background, I always understood that the "digital stair step" was a myth, but this video helped further elucidate that fact. Nice job.
There are advantages to 16 bit 192khz sample rates, That is in music production. This is because If you want to time distort an audio track or have it slow down (vinyl effects) It will sound better when it's running at a quarter of the speed because it won't sound so compressed and low frequency. Because all of the ultrasonic frequencies are also being brought down to within audible range so, of course after you do all of your editing, the final production can be 44.1 of course, but for production reasons, it's the equivalent of using vector or extremely extremely high resolution logos, because depending on where you move it and scale it and make it really big or small, it might affect it more or less.
How much of what you do has a dozen or so audio sources? I'm just wondering. I was part of a group called the suspicious cheeselords and there were about a dozen of us. Recording myself I could never tell 44100 from 96000, but with all of us I could pass the double blind. I was wondering, do I have a bad DAC (behringer 1820) or am I actually losing directional data at 44100 through low detail making many overtones/different phases too hard to separate?
As a hobbyist, this video just made so many things make sense that id not understood before. Thanks!
Haha! I was about to comment that higher sample rates and bits are very important in music / audio production, but you finally told this at the very end of your video.
Great video btw!
The benefit of high bit depth i see is that you can record a quite sound without using the full amplitude range of the recording and after amplifying it's still capable due to the S/N ratio. If you had 16bit but had a quite sound using only 8bits you wouldn't do much with it as amplifying would increase the noise also. Having 24bits can be usefull when you dont use the full amplitude. You can crank the volume up and still get a decent sound with the S/N ratio low without interfering with the music (speech, any audio indeed).
Absolutely, for recording you want to use 24-bit and possibly a higher sampling rate.
@@gblargg Yaaaay, suddenly resolution matters in production. What a surprise.
What knowledge this boy has 😮 and the fastest growing channel of audio community.
I started following him when he had 77k subs now he is already on 200k+
I learn so much from him.
Thank you for sharing! ❤
Thanks for sticking with me for so long, DJ Suvy! Glad to help.
@@AudioUniversity No problem mate. Your knowledge is priceless. Keep up!
@@AudioUniversity Have you examined the myth of mp3 compression?
@@rafaelallenblock That's a video I'd like to see!
@@rafaelallenblock agree, let's talk MP3 quality.
As a signal processing student and an audiophile - I thank you. Very nicely explained.
Students trying to flex😂
@@ClosestNearUtopia where in my comment do you see a flex? I'm expressing my appreciation for a good explanation, sounds like you're the one trying to flex here 🤨
This, and one other equation, was the only thing in my four-year electrical engineering degree I'd never heard of anywhere else. Thanks for making a video of it. When my professor proved in calculus that a perfect reproduction of a continuous wave exists in a jagged-looking digital stream, I gasped "what!" and the rest of my class looked disappointingly unmoved by the revelation.
It’s the magic of the Fourier transform.
How does this work? What if instead of fifty samples clearly visually suggesting a sine wave, I just have two samples? Is there somehow a way to reconstruct the original even then?? I don't know calculus so I know this may be a hard question to answer.
@@krisrhodes5180 Trying not to be too jargoney and also hoping I remember it correctly... "the Fourier transform of a rectangular pulse" can be applied to the concept of digital sample playback if each "stairstep" is transformed individually... the sinc functions overlap and the resulting waveform looks more and more like the original as you add all the harmonics of the stair steps.
He used ancient overhead transparencies to demonstrate it but I'm sure it could be done in Excel or MATLAB.
With filtering you can certainly approach the original waveform if it's a relatively simple combination of sine waves.
Fun fact: PVC molecules are big, and vinyl grooves are small. Together, these result in an equivalent dynamic resolution of 12 bit. Under ideal laboratory conditions.
But yeah, vinyl is "analog" as opposed to "digital"... that's just words and concepts. In physical reality, every 16 bit delta-sigma converter from the mid 90s is superior by orders of magnitude.
There are vinyls I like. And universally they are of medium to large groups, not soloists. CD is just better when it comes to signal to noise. I just wish that CD were higher sample rate for the orchestras.
It’s the AD that matters. Higher bit depth and sample rate, would be more important on that end, during capture of the sound. There is rounding and/or truncating happening there. It might sound better at lower, but it is a closer capture of the signal at higher rates.
Thanks for taking Monty´s great video to the "next generation". Used the original a lot in hifi-forums, not sure it helped, because many audiophiles share flatearther-vibes. But still one of the best explanations.
If I mixed in 16bits, no one complained or even figured it out. I bet no one cares if the track is cool and has sufficient "musical nutritional value" in it xD People seem to care more about the content of the music than the quality or bit depth. Even if a top track is released on old tape, people celebrate it more than a track that doesn't touch them and that in 24bit. At least that's how I feel xD
That's a great point, 4N4LOG! The most important element is the music itself.
However, there are some reasons why you probably should still use 24-bit audio for recording and mixing (when possible). It's not a deal-breaker, but it definitely helps during the recording and mixing phase! Playback should be 16-bit.
And then there's lofi hip hop that trying to emulate average/below average tape sounds yet enjoyed by many
I remember reading that even the best audio systems in the world have the noise that equals merely 18 bits one(so no idea why 24 bits exists at all).
And unfortunately there are people who just don't believe Nyquist-Shannon sampling theorem no matter what, they somehow hear the difference...
Thank you, I now understand what I don’t like about my nephew’s band’s music. It’s as I thought, the sample frequency is to low. They’ve been compensating by over processing the signals.
A long time ago I wrote a program to reduce the bit depth and sample rate of audio recordings so my students could hear what difference it made. One of them had just bought a CD of Black Sabbath's Greatest Hits so we gave it a try with my program. However, Ozzy Osbourne's vocals sounded exactly the same unless the bit depth was reduced to 4...
When I saw the thumbnail, I was like "yup, that's what the LPF slightly below the Nyquist frequency is for, don't people know?"
And then it hit me. No, people don't know. Great video, thanks!
The audio quality is quite sufficient for the CD format, it reminds me when I went from Vinyl to CD I was especially impressed by the dynamics and no background noise... The high fidelity is there remains to choose good speakers that will define the timbre of what we listen...
Vinyl still sounds better even with it's flaws
The problem with this simplistic experiment using a single frequency, continuous sine wave is that it completely eliminates any discussion of phase shift.
Go to a high end audio store and listen to some music recorded on a CD that is played back on a player with 4 and 8x oversampling. The difference will astound you.
Why? Oversampling permits the use of a lower order filter to reconstruct the analog waveform from rhe samples. What you don't hear discussed in the video is that filters (that reconstuct the signal) have varying phase shifts at different frequencies. Phase shift is time delay and that in turn makes the tones you hear move spatially in the audio you hear.
The closer to Nyquist the sampling rate the sharper the filter rolloff needed for reconstruction which in turn necessitates a higher order filter with, you guessed it, a much more distorted phase response.
So while 44 kHz audio meets the Nyquist criterion to recover the frequency content, it won't be sufficient to get all the phase info you need for crisp sounding stereo audio. Oversampling is critical for good audio.
Don't believe me? Like i said, go to a high end audio store and hear for yourself!
Oversampling in the DAC is digital interpolation between the recorded samples. It doesn't add any information. It just means you're doing some of the filtering digitally and the rest analog.
all modern audio ADCs and DACs use linear-phase FIR filters for anti-aliasing
12:23 if you know how loud a 90dB sound is, a 120dB sound is 1000x louder than it (as the decibel system is an exponential system, +10dB is equivalent to 10x the volume)
As a format, CDs offer twice as much dynamic range as even vinyl, but it's a shame a lot of music made today is just not mastered using any of it.
The dynamic range of vinyl was at best about 70 dB and CD is 96 dB or at the most 120 dB with dither, but yes, the last 40 years has been ruined by the "Loudness Wars". So sad. I haven't bought any music in the last 20 years.
The same could be said back in the day of analog recording. Hi-fidelity is usually saved for classical, with most pop and rock being recorded with compression play-back in mind, ie, AM/FM radio.
@@garymiles484 you're missing out, tons of less well known artists are producing beautifully mixed and mastered music. tired of hearing old people complain about music these days lol you all just aren't looking and listening hard enough
@@DAVID-io9nj It was necessary to use compression to make it audible in a car over the racket made by the engine.
@@SublimeSynth I am an old fart who started listening to music in the late 60's. Just because the recording process is great does not mean the music is great. Lucky for you to find new music you like. I had given up on new music until I came across rock from Japan!
The simplest way to bust a stair step myth is to discover some of the maths included in signals. With a system that has just enough bandwidth to play 20kHz signals (analog low pass at 22kHz or so) you are just not able to produce this kind of stairs because replicating those stairs would require huge bandwidth way beyond the signal played. So when there are samples every 44100th of a second, the "stairs" simply are not going through the low pass that should always be there after the conversion.
Exactly, the sad part is that mention of the filters was left out of the original video, I think for pro-digital propaganda reasons.
Without the anti-aliasing input filter and a extraordinary microphone a strong 30kHz signal (consumer ultrasonic cleaners are often in the 25-37kHz rage) would alias to 44.1-30=14.1kHz and create significant distortion in the digital steam from a noise source that no human ear could hear.
Without the output/reconstruction filter the 44.1kHz square wave component of the very real stair step wave would result in lots of wasted power in the output amplifier stages and perhaps overdrive any ultrasonic capable tweeter units but still remain inaudible to humans unless it caused clipping or other non-linear distortion.
@@KallePihlajasaari It was left out because it's an implementation detail of how a DAC works, which he didn't discuss because it isn't important for the topic of the video. What matters is that you can reconstruct a bandlimited audio signal from its discrete samples, provided the sample rate is more than twice the bandwidth, and a modern audio DAC is able to perform this reconstruction with excellent fidelity. How this works is interesting too, but not needed for a conceptual understanding of digital audio and its properties.
@@MatthijsvanDuin The lollipop representation is even less needed for any understanding because it is simply a graphical representation of the digital data stream and it NEVER appear anywhere but on paper and never in the output and not even really anywhere in the real world unless someone was looking at the differentiated output of a radiation detector, which is avoided because pulse height is usually integrated so it can be measured with more accuracy and ease. Look at the clickbait thumbnail of the video, this is propaganda and it is a crying shame someone has to repeat/reinforce it to pander to new generation of confused digital audio acolytes.
Digital audio does what it does because of mathematics and engineering, not because of smoke and mirrors.
This is one of those videos I hope I come back to in ten years and completely understand. Like, I know there's some wildly smart and mind-blowing stuff being talked about and shown, but most of the video goes over my head. Still, it's fascinating to watch. Thank you so much for putting this content out for people to watch! I can't wait to fully understand this!
Thank you so much! This was extremely informative. I've always wondered about this, since I couldn't hear the difference between "CD" quality vs higher bit and sample rates. I also enjoy all of your other videos! Kind regards!
RIGHT ON! 😎👍👍 Being old af, we learned all this when digital audio was first hitting the market.
It’s kind of a shame that you still have to walk people through this.
I think people are confused by the higher data rate that computers often need to PROCESS audio. This is a completely unrelated function.
Music is not a single sine wave of invariable pitch and volume. Thus you were fooled by this video. Sorry. In addition, bits don't equate to dB. Bits simply define a resolution range. You can use 16 bit to define 65536 different volume levels between 0dB and 1dB, or 65536 different volume levels between 0 and 1000dB. So here too you were fooled.
@@Äpple-pie-5k watch the video again and this time use your brain
Good and informative, (old CS/EE engineer with audio interest here). Remember back at uni in the 90s we had labs doing exactly this, learning about different types of wave reconstruction from digital data in DA-converters (there is more to it than the video mentions but the conclusion is very valid).
One thing that is not mentioned with dynamic range is that there has been historically hard to have a standardized level for where to put the mean audio level in recording digitally. This is what is called the "loudness wars" where listeners tend to like music etc at higher volume. Also the reason commercials are usually at a way louder volume. This drove recording industry to record at higher and higher levels.
So a lot of CDs actually had a very high average volume meaning that it was less dynamic range to use. This made them sound inferior to what the CD-standard actually allowed. A lot of CDs were "badly" recorded so to speak. Ideally you would want the digital recording to be fit within the dynamic range in a way that maximizes the format.
The effect of reduced dynamic range is similar to how "nightmode" sounds.
This is one reason the classic vinylrecords can sound better than CDs. The format in itself is wildy inferior but they were mixed a lot better.
There is also another issue related to this and that is the intentional compression of dynamic range to sound better on devices that are not able to accurately reproduce sound.
There are some very interesting soundclips comparing speech and music with and without artificial dynamic compression. This is a very big area with lots of opionions (some mixers feel the compression is part of the style).
Personally I want my audio to have as much dynamic range as possible. To me it sounds a lot better, subjectively there is more texture in the music.
@@jim9930 Cool setup :) Yeah recording technique makes a huge difference. But with compression I’d did not mean compression as in bit/bandwidth reduction but as in compression of dynamic range. You hear it is a lot of modern recordings. If someone is singing it is the same volume regardless if they are belting or if they are whispering (a bit exaggerated but not too much). This is done in software usually for aesthetic reasons and for it it sound better on poor devices (cellphones, laptop, iPad etc etc).
Have listened to some horn systems and they have immense directivity and effectiveness. Tend to get very very big if you want to do good bass with them. Personally I have tinkered a bit with something called orthoaccoustic speakers that utilise room reflections more to create a more enveloping soundscape.
Watched Monty's video right at the beginning of my journey into digital audio and video. Really solid and great deterrent for a lot of the audiophile nonsense lol
Thank you so much for making a video about this topic. It was high time, that someone with the proper reach does a video and debunks these myths. We really need videos like this! It’s just a shame, that nobody of the reviewers, who keep „selling“ a high sampling rate or bit depth on audio interfaces, will never really care to watch this video.
Problem is they CREATED far more false myths than they debunked. The only "debunked myth" was using the stair-step as an [over]simplified way to visualize how limited sampling rates always lose data. The actual mathematics for how sampling loses data is more complex. It comes down to using mathematical guesstimates in the reconstruction filters during the D-to-A playback. But didn't they just disprove that? No. Real music isn't the simplest wave theoretically possible--the 1kHz sine wave. Absolutely No. Real music is a true ocean of multiple waves and overtones with ever-changing apex and trough points which are escaping the sampling rate and have to be "guessed" by the reconstruction filter algorithms. These algorithms create all manner of "false local minima" and "false local maxima" -- basically you might call them hallucinated soundwaves artificially interjected into the playback.
Reconstruction algorithms therefore introduce false and lossy artifacts during the D-to-A playback. What Nyquist can do for a single invariable sine wave is true science, but only an idiot mistakes that for variable waves some of which aren't sinusoidal.
Moronic and utterly sophomoric and just-plain-dumb-as-f\/ck is the ignorant ineptitude of using a 1kHz sine wave of unchanging volume, as a proxy example for recorded music. Real music presents mind-numbingly difficult mathematical and engineering challenges which lower the quality of all digital playback at ANY AND ALL sample rates currently used by humankind. Want proof? Until you can hear digitally recorded cymbals sound like real metallic rainbow shimmers rippling through a 3D holographic room--which you can't--you have LOSSY LOW FIDELITY. 50 years from now they'll laugh at the arrogance of those saying we can't surpass 16/44k1. At 44k Hz sampling, cymbals sound like some someone pronouncing "ch ch ch" dubbed over the sound of shredding paper.
In 1984 Wendy Carlos released the album digital moonscapes, which was created with digital synthesis, and recorded digitally.
Dire Straights 1985 album Brothers in Arms was digitally recorded.
There was extensive testing done at that time sending analog audio through digital conversion and back. it was determined that literally no one could tell the difference between the converted signal and the original. In my mind, this has been settled for 40 years.
FINALLY !
Someone besides me finally GETS IT !
This is possibly the best explanation I've ever heard on the topic of digital audio.
And I've been doing this long before there was such a thing (62 years and counting.)
And yes, 96 dBs of RMS dynamic range is far more than enough for any reasonable playback.
I typically mix and master at 80 dBs SPL (at 3 ft.) and that's plenty loud enough.
Once again, excellent presentation, thank you !
Bill P.
Claude Shannon and Harry Nyquist understood it a good century ago.
@nrezmerski Replace "good" with "approximately".
And here I am at 71 years old and I can barely stand the 25kHz scream of the dentist's scaler. Those things are LOUD if you can hear them. As I age I'm losing my high-frequency hearing, but I can still hear these supposedly quiet machinex. One of my sons has the same range of hearing, and can't have the scalers used on him. Silent Dog Whistle? No such thing as fas as I'm concerned.
As for digital music? The problem I experience is the lack of dynamics. MP3 is the worst thing that ever happened to music.
I wish i can hear 20k again
Maybe you just hear the subharmonics when you are at the dentist. As to MP3, read the research by Plenge, Kürer, Wilkins, who came up with the mathematical model of the ear in the 70s, i.e., they analyzed what the human ear can hear and what not. The people at Fraunhofer then took their results and used them to create the MP3 standard...
10:27 That's only on Type 0 tapes. Quality Type 1 tapes definitely don't sound like they have 5-6 bits of S/N ratio, as shown by VWestlife and Cassette Comeback, when the level is adjusted properly. They are closer to 9 bits.
And the hiss can also be mitigated quite a bit with those mains powered bulk tape erasers, if your tape deck doesn't erase it properly.
Very interesting, thank you. In my youth I had a hearing range (like you said), of 16Hz to about 19kHz, and an 'awareness' up to 25 kHz. I now have 67 year old ears and a very reduced audible range, so my 30 year-old record/CD/radio setup is quite adequate and still sounds as good as it did when new.
5:29 this is only true if there are as many or fewer waveforms than sample points, which of course will always be true when there's a single sine wave. I'd like to see this experiment replicated with 20 different frequencies.
Proper, imperical, practical analysis….. videos like this are worth their weight in gold. Well done!
One effect not explored in these experiments is the phase of the signal. As the pitch of the signal approaches the Nyquist limit, while you still have the frequency itself preserved as a pure tone, its phase will be more and more forcibly synchronized to the samples. This interferes with binaural listening for which we use stereophonic sound reproduction. This would be why it would be that played from a CD, while a violin or a trumpet would be easily localizable to the human pair of ears, a cymbal, full of extremely high frequencies and transients, will tend to sound as if it is coming from all over the place, and not at all as if it comes from where it had in fact been located. True pure analog recording does not have this problem.
I think 400Hz is the frequency where binaural hearing changes from phase to amplitude detection. At 10kHz, for example, the wave length is 3cm, and the ears are too far apart to detect a phase difference, so the brain uses amplitude differences to localize the sound. At least, this is what I seem to remember from Dr, Plenge's course on the psychology of hearing at TU Berlin.
No. Frequency, amplitude, and phase information are all accurately preserved. If the video had used a cymbal crash (limited to 20 kHz) rather than a sinewave, the input and output waveforms would still match (assuming the digital processor box has good filters).
It would be interesting to see what a non-sinusoidal signal turns into with this conversion and how much noise there will be in the final signal. In the above example, an excellent sine wave can be obtained from a sine wave simply because this is how the digital-to-analog converter works.
This is exactly what I was thinking. If by default the DAC is smoothing the waveform assuming a single frequency (clean sine wave) then when you modulate some additional frequencies in there may be a lot more difference between the input & output.
Any non sinusoidal waveform (sawtooth, square) basically everything with "corners" has a infinite frequency spectrum, and is therefore beyond 22kHz.
@@MatthiasSchaffSo what? Is sawtooth or square signals not audible? Or no visible on osciloscope? But such signals will definitely show how analog-digital-analog transformation works with all features. Clear sine is not a good example to show that there is no loss with such transformation of signal.
@@MatthiasSchaff the key is using something other than a single frequency. This could easily be established by modulating several frequencies together for a comparison. If a DAC is designed to always smooth as if it is a clean single frequency sine wave then a lot can be lost.
@@GardeningSolutions Exactly. And it so happens music usually is made up of many frequencies all at the same time ! Why this is absent from the discussion here except in this subthread makes the mind boggle. Not to speak of unpitched sounds !
I never even considered the idea that captured samples can only result in one possible output. That revelation is kinda of a game changer for me. Great video!
But that's the point in the video where suddenly there's no explanation where one is needed. :/ How can there only be one possible output? He shows someone drawing multiple possible outputs, then just says "turns out, nope, you can't." Why not? What if there had only been two or three samples? How could there be only one possible output then?
@@krisrhodes5180 if you draw a shape that is clearly wrong but hits the sample points, the you have tried to capture something that has way too high of a frequency for your sampling rate. Thats why sampling frequency needs to be double of the intended frequency range.
Okay I think I may be starting to see how this works. If I have just two samples, then the highest frequency this works for would be one (over the same time as the two samples). I can imagine that if I'm told "these two samples represent a sine wave at a frequency of one or lower," maybe there's a unique solution there. (Still feels like there should be a solution at every possible frequency one or lower but I can believe that feeling is wrong.)
@@krisrhodes5180 It's called the "Nyquist-Shannon sampling theorem", and I think if he went into the math in his video he would just lose the audience.
I said it before, but I am going to repeat it here: these videos of yours are priceless when it comes to bring facts into discussions around audio restitution. So many unscientific arguments are thrown at people's faces in audiophiles forums that debates often end up becoming of an almost religious nature. Pedagogical ressources like yours are therefore highly needed and most welcome.
To be frank, though, equipment manufacturers are also to blame for this. Originally, these discussions indeed sprouted from a small portion of the audiophile community who had a grudge against the CD format (for reasons I still fail to understand). Once that debate took off however, many manufacturers of consumer audio equipment started playing into it by feeding people's demand for pointlessly over-sampled signals or obsolete analog devices disguised as "high-end" audio equipment. It all felt very much like how, at about the same time (in the late 90s and forward), computer manufacturers would focus their public communication on the high frequencies of their processors, while integrating those in motherboards with otherwise sluggish buses (making said high frequencies totally inoperative).
Now, if I may make a suggestion:
You should at some point make a video explaining the technology behind Class D amplifiers. On the topic of debunking myths about "jagged" signals in digital audio, one could hardly find a better counter-example than amplifiers who literally can only spit out ones and zeros on an audio line, while still allowing faithful audio reproduction.
Just my two cents.
Thanks for the kind words, Going Modular! I'm glad you find my videos valuable. Also, I'll add Class D amps to the list of topics to cover in future videos. Thank you!
let's also not forget that old "digital" storage media do have degradation unlike modern flash memory or hard drives with all the software making sure the bits are in the right place. early digital formats suffers from data corruption and it ruins the audio result despite how good the decoding part is.
@@fltfathin It was true to a point for a few recordable media like the early DAT format for instance, but that concerned practically few people, and only in the professional range. The dominant consumer digital non-inscriptible format, i.e. the good old CD, was (and is) pretty much flawless (unless mechanically corrupted).
On the other side, people do not realise, quite often, how rapidly analog recordings on magnetic tapes used to degrade. Most of the time, unless the storage conditions were optimal and the tapes of high-end quality, signal fading and cross-talking between tracks would become critical after 10-15 years. Not to mention the physical degradation of the tape itself, which sometimes came first, literally making it crumble to dust as it was played back.
And of course, not all studio had the space (or the will) to simply store most of their productions. Most pre-mixed multi-tracks recordings are lost (but then, that is also true of digital recordings on physical media _ like the DAT, again _ when Megabytes used to cost good money and take storage space).
Very good video. But...
The professor said it -- distinctly and deliberately -- then glossed over it: "Band limited signal." This means that the ONLY waveform the CD is capable of reproducing without distortion at 20kHz is the sine wave. Feed the system a square wave (which theoretically requires an infinite bandwidth), and you get a delightfully pure sine wave out, which is an egregious distortion of the input. Granted, if the human ear is also band limited (can't hear over 20 kHz), then the perceived reproduction is perfect!
I'm not complaining about digital (love it), but this fact should always be mentioned in a discussion of the topic.
Any signal is a sum of infinitely many sine waves. What matters for human hearing is those sine waves. Watch "The Other Square Wave" by monster860.
Very interesting video, but we were only looking from the current standard to higher frequencies and bit depths (I know, there was also 8 bit).
What I would find way more interesting would be at which point the signal will get distorted if we lower the sample rate and/or bit depth. Obviously,
if you would use only 1 bit, the signal would not be presented very well, but at what point does the distortion begin?
Coming from an EE background, I find it amazing, nay, almost scary, that these myths still persist. Thanks for a great video!
I believe the 44.1 kHz audio used by audio CDs, Was really because at the time with Sony PCM, that was the maximum that could be recorded as data on U-matic videotape for PAL and NTSC standards. And later on, Betamax was often used for digital recording distributions to the CD pressing plants. This was before DAT became the norm in the recording industry.
Ahh penny dropped moment. I took a tour of a CD pressing plant in the 90's when it was all quite new, and we were shown the process from start to finish. When the engineer showed us the source material as a videotape we screwed our faces up, I guess thinking it arrived in analogue form. As an aside, they had to destroy a large proportion of the pressed CD's as they had spots in them, usually black. They played fine, but consumers would think them faulty and return them. Another thing they showed us, and gave us a sample of which I still have today, was the ability to press a hologram into the disk instead of printing the tracklist or whatever. Looks amazing, but I have never seen it used on a CD release.
I had absolutely no idea I would enjoy being informed this much. Thank you
It's all about context. Which is also skipped over here a little bit. On it's own, yes 16 bit DNR is enough. Unfortunately we don't use a DAC on its own, there will be an amplifier behind it. This amplifier also amplifies the noise. Just in a every day home setting this is often not a big deal. In a very quiet (mixing) environment, speakers that have active filters and very high efficiency high frequency units, the noise can be easily be audibility. The same goes for a big line array system with like 24 compression drivers with a sensitivity of 108dB/2.83V each. How do I know? I develop these systems and I can tell from 15 years of experience that in both cases those noise levels are clearly audible for every human being. There is also a good argument of having some safety margin. Although that's more on the recording site. When music is mastered, there is a lot of EQ'ing going on, you also don't want to clip your signal. Both combined will already give at least a 24-32dB penalty on your DNR. Resulting in an effective SNR of just about 75dB or so. Similar things can happen when it cones down to sample rate.
So while I agree with the message that people shouldn't be so worried about DNR/SNR and samplerates as well as debunking some myths, this video still gets a thumbs down from me because it's extremely over simplified. It doesn't explain why some of those higher numbers are actually very useful. Therefore also spreading misinformation.
An amp's job is to amplify all frequencies equally. The db's will stay the same while the dbm's will all increase -- there's no reason why an (excellent) amp should amplify the source noise more than the source signal. (the amp will add its own noise, to be sure). You're totally correct about the safety margins -- that's why we master and digitize at much much higher precision/frequency than the output. And yes, I am yet to see a speaker that can reproduce an even 80db range without distortion 🙂
@@dmitryjoy That's true, the issue is when there no signal present, it will still amplify the noise. Resulting in audible noise at quiet moments for example.
Which is just a small annoyance at like monitor speakers, or can be very disturbing for a professional musician with an active floor monitor (with a compression driver) during a concert.
But during just regular play at max volume, yes you're correct.
Very impressive content. This is a topic I am well-Informed about, and a misconception that I’ve explained to people on a few occasions. You did a fantastic job of explaining the fundamental fact that the PCM transcription can only be decoded into an analog duplicate of the analog input signal. Noise floor and headroom of the analog components in an ADC and in DAC for playback will matter if you are interfacing with analog audio equipment, so gain-staging and phase and the usual analog caveats still apply- and there absolutely have been very poorly-implemented examples that set bad examples of what digital audio ‘sounds like’- .mp3 stereo 44.1kHz at 128kbps was the norm for a long time, and some .mp3 codecs performed very poorly at properly encoding high frequency content, there is very obvious aliasing that can be heard for everything above 13.5khz.
I clicked on this video mainly to check and see if you had your facts straight, and man you sourced an excellent video explanation also. I did not get to this understanding of the PCM wave function on my own studying until I was about 40 years old, after spending about a decade as a telecommunications technician and getting to the point of telecom engineer. I’ve been making music with computers and instruments 30+ years, studying recording technology and techniques, and building my own synthesizers- this level of technical detail is satisfied that you didn’t gloss over any important points and that you drilled down to the absolute truth of the matter without resorting to an analogy (pardon the pun) or a metaphor to explain why.
I was on the edge of my seat waiting for you to explain why higher sample rate/Nyquist is desirable in digital audio production, and might be an interesting/informative counterpoint to mention that some production choices intentionally lower the sample rate/bit depth and collapse mono in order to induce digital artifacts during conversion, such as aliasing and quantization loss (the Lo-Fi phenomenon, 12-bit samplers and wavetable synths making a big comeback, the proliferation of bitcrusher FX, etc)
Yes some codecs were dreadful. I remember satellite TV undersampled their transmissions on many programs in early years and sound scores had nasty and painful artifacts. Yamaha's TwinVQ codecs were very good, but all suffered with bad artifacts when the sampling was cut too far.
A couple years ago I re-released an old album of mine re-mastered from original tapes. Lossless 24-bit FLAC is available, but people keep asking me "when are you going to release this on vinyl?" It would sound worse on vinyl, especially since the album is 47 minutes long. It was quiet on the original vinyl with bass turned down by George Horn in mastering to keep needle from skipping. It would probably have to be 2 or 3 EPs, 45 RPM 12 inch to come close to the listener experience of the CD quality (16 bit) digital files.
What concerns me with these experiments is that they are done on pure sine waves. Audio is rarely a pure sine waves. High-frequency (including in-audible) harmonics are additive. So a bit depth helps to re-claim some of these harmonics for a purer sound. It's not necessarily only about the noise floor. It's also about recreating the harmonics on the signal. But it is a good demo none-the-less.
That being said, most DAC's nowadays do not do the stair-step method, they climb the voltages between the points so you get more of a "lined" slope. instead of a histogram looking thing.... Modern DAC's are amazing, and represent the MPEG-destroyed digital audio signal to it's very best!
I feel like I learned so much in such a short time. I couldn't ask for anything more. Perfectly concise.
I can think of only one purpose for high sample rates; recording with the intent to slow down or stretch the audio while preserving quality. Does someone more informed than me have another take?
All true, and thanks for an excellent video - it explains it very well!
16-bit/44.1kHz really is enough, despite what some so-called audiophiles otherwise believe (or pretend).
i don't even bother mastering at hi-res because i don't think the audiophile crowd will particularly care about the music i produce, let alone the ones who simply want to enjoy the music (this is where we should be prioritizing about, anyway).
Good point, AI R! Plus, in my research for this video, I found that 192kHz audio will probably have even LOWER fidelity when played back on most systems. The intermodulation of the ultrasonics actually creates audible distortions! So, a true audiophile wouldn't want you to master in high-res formats anyway, I'd say! Read more about that here, if you're interested: people.xiph.org/~xiphmont/demo/neil-young.html
@@AudioUniversity So could I use that intermodulation of the ultrasonics to make a different kind of distortion than what normal effects give?
@Jaakob Aakko It’s a subtle effect, but the link above has a few audio examples.
Two arguments: First, we don’t listen to audio tones. Recorded music is nothing close to a sine wave. Live music is represented by complex fundamentals and overtones. That’s how we learn to differentiate between a symbol crash and an automobile crash or between an oboe and a clarinet. Comparing a DAC to analog with pure audio tones is something else entirely. Second, if one grows up never hearing a purely logarithmic recording, it wouldn’t be possible to realize a difference actually exists. Although, digitization brings about affordable,and compact reproduction equipment at an order of magnitude less in cost compared to a comparable analog system, it also comes at a price.
Finally setting a music CD as the ultimate goal to analog, Roy Orbison said it best”Only in Dreams” unless pure sine wave tone recovery is the goal.
Anything in between introduces coloring. At best, a perfect digitization can only “approach” the complex sound we hear.
I worked for Harman in the mid 90’s & we did blind tests at 44.1, 48 & 96k. Everybody could hear the difference between 44.1 & 48 but most couldn’t hear any difference between 48 & 96. I do all my work in 48k now
People don't mix in 96k for the sound but for the latency. Just as you dont mix in 32 bit for the depth but to reduce noise input from mics.
"everyone could hear difference between 44 and 48"? no they didnt...
@@matswessling6600 believe what you want and stay mediocre
@@GroverTD LOL. You should do your work in 96k and 32b to avoid artifacts, and distribute in 44 k, 16 bit.
Learn som facts and dont do stupid things.
But that's about the quality of processing after DAC. A higher sampling rate allows for poorer post-processing quality.